Unmanned Aerial Vehicles (UAVs) have recently emerged as a key component for various 5G applications, offering either precise positioning or high-speed, low-latency data communication services. However, delivering both wireless positioning and data communication at the same time using UAVs remains challenging due to the different requirements of each service. While data communication demands high throughput, positioning services necessitate the establishment of multiple connections simultaneously, but achieving high throughput for all the simultaneous connections remains challenging. Pattern Division Multiple Access (PDMA), which flexibly shares Resource Elements (REs) among multiple users, presents a promising solution to this issue. In this study, we investigate the joint provisioning of communication and positioning services in a UAV airborne network using PDMA. We propose a comprehensive approach that addresses the joint problem of user-to-UAV association, RE allocation, and transmission power control. Our goal is to enhance the precision of positioning services while satisfying communication constraints. To tackle this problem, we develop both an exact solution and a Multi-Agent Deep Reinforcement Learning (MADRL) algorithm, tailored for this highly complex non-convex optimization problem. Extensive simulations demonstrate that our attentional MADDPG algorithm achieves higher positioning accuracy compared to state-of-the-art solutions and can efficiently address interferences, thereby improving both services.

Recently, companies have focused on developing new technologies for air mobility using flying cars to alleviate road congestion in urban areas. A critical aspect to consider is the seamless integration of flying cars with their ground-based counterparts in the 5G network, where ground-based cars can provide transit functions, including access to vertiports and urban amenities. Additionally, flying and ground-based cars require various services with different requirements, such as path planning, remote diagnosis, and autonomous driving/piloting. Supporting these services in 5G networks is challenging due to the high mobility and stringent network latency requirements. Network slicing can be a promising solution to meet these requirements. However, the literature lacks comprehensive research on combining flying and ground-based cars in network slicing, where resource under-provisioning can cause the violation of service requirements. We propose three-level closed-loops for sliced resource block management to satisfy the delay budget constraint of flying and ground-based cars while avoiding resource under-provisioning. We present a reward function and continual learning that links these closed-loops. Furthermore, we use Ape-X as distributed deep reinforcement learning to maximize reward and continual learning to improve resource allocation via prediction. The simulation results demonstrate that the proposed approach maximizes delay requirement satisfaction.

This paper addresses energy-efficient processing in the Open Radio Access Network (O-RAN) within the O-RAN Cloud (O-Cloud) infrastructure, focusing on optimal processor power scaling while scheduling packets under delay constraints. Traditionally, real-time packet scheduling tends to be suboptimal, as it prioritizes processing capacity over energy efficiency. To address this issue, we formulate a Mixed Integer Programming (MIP) model and develop a heuristic approach, along with a Convolutional Neural Network(CNN) for real-time scheduling. To enhance the CNN model, we introduce the CNN Binary Search Scheduling (CNN-BSS) algorithm, guided by the heuristic solution. Simulations show a 29.17 % reduction in energy consumption, significantly outperforming conventional methods.

The widespread adoption of IoT networks has made them vulnerable to jamming attacks, which disrupt communication and compromise critical applications. Traditional jamming detection methods face challenges due to resource constraints and the need for large labeled datasets. This paper proposes a Deep Active Learning (DAL) framework for jamming detection and classification, addressing these limitations by minimizing the reliance on annotated data while maintaining high accuracy. Leveraging pool-based and uncertainty sampling, our approach iteratively selects the most informative data points for labeling, significantly reducing the dataset size required for training. Using a real-world IoT dataset, the proposed framework achieved 98% accuracy with only 50% of the available dataset. This represents a 13% improvement in accuracy and a 50% reduction in dataset size. Experimental results, validated through Monte Carlo simulations, demonstrated the model’s robustness in distinguishing between normal channels, constant jammers, and periodic jammers, with minimal misclassification. The framework’s efficiency and accuracy make it a practical solution for resource-constrained IoT networks.

This paper introduces a novel approach to AI-powered digital-twins-assisted robotic control in automated warehouses, integrating the kinetic models of robots with real-time synchronization of digital-twins. The proposed framework utilizes Ultra-Reliable Low-Latency Communication (URLLC) over 5G networks to enable seamless interaction between the physical robots and AI-driven models in the cyber twin. We formulate an optimization problem aimed at minimizing energy consumption during digital-twins-driven robotic operations, thereby enhancing both operational efficiency and energy efficiency. A Deep Reinforcement Learning (DRL)-based approach is developed for the adaptive learning of the AI models in the cyber twin, facilitating autonomous simulation and real-time decision-making for efficient robotic control. Additionally, we propose a game-theory-based resource allocation strategy to optimize the distribution of computational resources for continuous and adaptive learning within AI models. Numerical results demonstrate that the proposed game-based resource allocation scheme achieves Nash equilibrium, significantly improving performance in terms of energy consumption and resource utilization compared to the state-of-the-art DRL-based resource allocation scheme.

Internet access in rural areas should be improved to support digital inclusion and 5G services. Due to the high deployment costs of fiber optics in these areas, Fixed Wireless Access (FWA) has become a preferable alternative. Additionally, the Open Radio Access Network (O-RAN) can facilitate the interoperability of FWA elements, allowing some FWA functions to be deployed at the edge cloud. However, deploying edge clouds in rural areas can increase network and energy costs. To address these challenges, we propose a closed-loop system assisted by a Digital Twin (DT) to automate energy-aware O-RAN based FWA resource management in rural areas. We consider the FWA and edge cloud as the Physical Twin (PT) and design a closed-loop that distributes radio resources to edge cloud instances for scheduling. We develop another closed-loop for intra-slice resource allocation to houses. We design an energy model that integrates radio resource allocation and formulate ultra-small and small-timescale optimizations for the PT to maximize slice requirement satisfaction while minimizing energy costs. We then design a reinforcement learning approach and successive convex approximation to address the formulated problems. We present a DT that replicates the PT by incorporating solution experiences into future states. The results show that our approach efficiently uses radio and energy resources.

Without network slicing supports, traditional Fronthaul architectures struggle to meet the demanding requirements of 5G networks, such as the ultra-low latency and high bit rate specified by the enhanced common public radio interface (eCPRI). In this paper, we design a novel Fronthaul architecture that leverages FlexEthernet (FlexE) over elastic optical network (EON) to enable Fronthaul slicing meeting 5G Fronthaul requirements. Our Fronthaul design is optimized by an integer linear programming (ILP) model, named eFFP, that minimizes the total cost of ownership (TCO). While eFFP meets the strict Fronthaul requirements by provisioning network resources based on worst-case traffic load, it tends to overestimate required bit rate as a result of the inherent uncertainty and variability in real-world traffic. To tackle this challenge, we introduce uFFP, a stochastic Fronthaul provisioning strategy tailored to accommodate uncertain traffic demands and mitigate expenditure wastage. Relying on historical data, uFFP assesses statistical characteristics of traffic patterns to better estimate Fronthaul bit rate. Subsequently, we employ chance-constrained optimization to reformulate the uFFP problem, which is approximately solved using a convex relaxation approach known as uFFPA, and optimally solved using a deep reinforcement learning (DRL) approach called uFFPL. Simulation results demonstrate that our proposed solutions achieve significant cost savings, reducing TCO by 39.79% compared to the baseline.

Open Radio Access Network (O-RAN) aims to establish an open and intelligent RAN architecture, enhancing flexibility, scalability, and network optimization. Machine learning (ML) technologies are pivotal in realizing these objectives by facilitating intelligent decision-making, automated optimization, and proactive maintenance. However, effectively selecting and deploying ML models within O-RAN to achieve energy efficiency poses significant challenges. In this paper, we propose a novel orchestration scheme tailored for next-generation systems, building upon and extending the foundational principles of the O-RAN paradigm. Our proposed orchestration policy offers a practical solution for deploying ML applications within the ORAN framework. Through comprehensive evaluation, our scheme demonstrates a remarkable reduction of up to 72.22% in energy consumption compared to the maximum performance baseline, while maintaining an accuracy level of approximately 94.56% relative to the same baseline.

Time series prediction plays a crucial role in optimizing the operation of communication networks. Applications of time series prediction include traffic prediction, channel state prediction, handover prediction, etc. However, training high-quality models for these tasks requires large volumes of historical data. This requirement may not be available in some scenarios. In this case, instance-based Transfer Learning (TL) comes as a prominent solution for this problem. However, a few concerns could be raised such as: 1) the time and bandwidth resources consumed in the transfer, 2) it will be hard to specify the amount of data to be transferred, and 3) in case of transferring a subset of the data, which subset is better to transfer. To address these challenges, we propose a novel approach for TL, which is similar to, but different than, instance-based TL based on generative models. We coined the new approach as Surrogate Data Source Transfer (SDST), in which a generative model is trained on the source task. We then transfer the model to the target task (with limited historical data). Extensive experiments confirm the superior performance of the proposed approach in terms of prediction accuracy and consumed resources (time and bandwidth). Our TL approach reduced the mean absolute percentage error (MAPE) by a margin that hits 81% in some datasets. For the source code and data, we refer to the repository https://github.com/MoeR3za/Korsahy_TGAN.

Recently, companies have increasingly developed new technologies for urban air mobility using flying cars to alleviate road congestion. Unfortunately, the seamless integration of flying cars with their ground-based counterparts in the 5G network, where ground-based cars can support flying cars in proving transit functions, has not yet been fully investigated. Flying and ground-based cars require various services, such as autonomous driving/plot, path planning, and remote di-agnosis. Supporting these services in 5G networks is challenging due to the high mobility and stringent network latency requirements. Network slicing is a promising solution. However, a comprehensive research on combining flying and ground-based cars in network slicing is still missing in the literature. Under-provisioning of radio resources can result in the violation of service requirements, while radio resource over-provisioning can cause resource under-utilization. We propose two-level closed-loops for Resource Block (RB) management to satisfy the delay budget constraint of flying and ground-based cars simultaneously while avoiding radio resource under/over provisioning. We design two closed-loops to map slices and services to Open RAN elements for radio resource scheduling and to allocate RB to cars. We propose a zero-touch RB adjustment approach and link these two closed-loops through the reward function of a deep reinforcement learning algorithm that optimizes slice resources in real time. Results show that our approach maximizes delay requirement satisfaction while preventing RB under/over-provisioning.

Using UAVs (Unmanned Aerial Vehicles) to provide either precise positioning or high speed, low latency data communication service has recently emerged for a new class of 5G applications. However, the joint optimization of data communication and wireless positioning in a UAV airborne network is challenging because each service has different requirements. Communication services demand high throughput, while positioning services require the establishment of multiple connections simultaneously. Pattern Division Multiple Access (PDMA) is an innovative communication technique based on Non-orthogonal Multiple Access (NOMA), which facilitates the sharing of REs (Resource Elements) among multiple users. In this paper, we formulate the joint problem of user-to-UAV association, RE allocation, and transmission power control, aimed to improve the precision of positioning service while meeting the communication constraints. We propose both exact and a Multi-Agent Deep Reinforcement Learning (MADRL) algorithm to solve this non-convex optimization. In addition, to address the problem of interference among users, we propose an attentional Multi-agent Deep Deterministic Policy Gradient (MADDPG) approach. Extensive simulations demonstrated that our proposed algorithm can achieve higher positioning accuracy than state-of-the-art solutions while serving more users.

In light of the essential need for persistent, real-time monitoring of natural environments, our research introduces an innovative approach that integrates a predictive solar energy model in optimizing solar-powered environmental monitoring efficiency with energy constraints. We address the challenge of solar energy variability by forecasting solar energy availability based on the Random Forest model. Then, we integrate the prediction ability into the Integer Linear Program optimization framework of the monitoring efficiency subjected to energy constraints. Experimental results demonstrate our approach’s effectiveness compared to traditional methods, underscoring its potential for enhancing sensor networks’ sustainability and operational efficiency in natural environments.

The rapid advancement of 5G technology has significantly increased energy consumption, underscoring the need for advanced energy management solutions. Proactive energy management, which relies on accurate predictions of network load to enable timely adaptive actions, emerges as a key strategy in addressing this challenge. In this study, we introduce a refined approach to forecasting traffic load in 5G networks, emphasizing the integration of statistical and temporal feature engineering. This methodology is aimed at capturing the intricate spatial and temporal patterns inherent in network data, thereby enhancing prediction accuracy. Leveraging an existing dataset comprising measurements from 1,000 base stations, we enriched this dataset with a set of derived features that reflect both temporal dynamics and load characteristics. Utilizing this enriched dataset, we trained and validated a suite of predictive models. Our findings reveal a notable improvement in the accuracy of traffic load predictions, outperforming standard baseline models. This underscores the effectiveness of our feature engineering approach in refining the predictive capabilities of models, paving the way for more efficient and proactive energy management in 5G networks.

The development of beyond-fifth-generation (B5G) communication systems introduces challenges in maintaining timing and frequency synchronization, especially in low SNR and extended coverage scenarios. Accurate carrier frequency offset (CFO) estimation is crucial for establishing calls under such conditions. Existing methods, like maximum likelihood estimation, have limitations, while machine learning (ML) techniques have shown promise in wireless communication. In this work, we propose an ML-based approach using Long Short-Term Memory (LSTM) neural networks and automated machine learning (AutoML) to tune hyperparameters and improve CFO estimation accuracy. We compare our model with a gradientboosting machine (GBM) approach and demonstrate superior accuracy. Our research addresses CFO estimation challenges in B5G systems and offers valuable insights for the development of robust techniques in advanced communication systems.

Traditional FlexEthernet (FlexE) defragmentation schemes have successfully been employed to reallocate the slots of affected FlexE clients during network changes such as FlexE physical link (PHY) failures in multi-layer multidomain (MLMD) networks in the context of fixed traffic rate. In such a context, constant slots of FlexE clients are statically pre-assigned using a round-robin algorithm before the network change takes place. However, in more realistic scenarios where traffic varies over time, this static assignment requires multiple defragmentation steps, potentially violating the maximum tolerated reconfiguration time and resulting in traffic loss. Therefore, a dynamic defragmentation scheme is required to efficiently move the affected slots without disrupting unaffected traffic. This paper introduces FDL, a semi-supervised learning approach designed to efficiently address the FlexE defragmentation problem under time-varying traffic conditions. FDL leverages an autoencoder for unsupervised pre-training, particularly due to the considerable amount of unlabeled data resulting from the unsolvable high-complexity optimization problem. To optimize throughput while adhering to reconfiguration time deadlines, FDL employs a gated recurrent unit (GRU) structure to forecast the future reassignment of FlexE clients’ slots over the defragmentation steps ahead. Simulation results demonstrate that the proposed FDL achieves a throughput that is 17.18% higher than a state-of-the-art approach.

The 6G transport network facilitates the Internet of Everything (IoE), carrying numerous services and emphasizing the paramount importance of its reliability. However, within the transport network, the issue of co-route fibers arises. The co-route fibers, encompassing both co-cable and co-trench fibers, presents a significant latent hazard for service disruptions, posing a substantial threat to the seamless connectivity envisioned for the 6G era of pervasive IoE. The segregation of communication and sensing in the transmission network results in mutual interference between communication and sensing signals, rendering it difficult to promptly address sudden fiber interruptions. This article proposes an integrated sensing and communication (ISAC) architecture within transport networks, aiming at the online discernment of co-cable fibers, characterization of fiber optic trenches, and real-time classification of fiber vibration events. In the domain of co-cable fiber identification, our approach has successfully reduced the nuisance alarm rate to an impressive 5.3%, while simultaneously elevating the recognition accuracy to an outstanding 99.7%. As for co-trench fiber identification, our proposed methodology not only facilitates the discernment of co-trench fibers but also achieves an impressive accuracy of 97.7% in classifying fiber trenches. Moreover, in the realm of fiber state prediction, our solution has achieved a remarkable recognition accuracy of 98% across six distinct vibration events. These results underscore the robust performance of the proposed ISAC architecture, which will effectively safeguard the survivability of 6G IoE.

In the realm of emergency disaster relief, it is paramount to attain a thorough comprehension of the semantic information associated with the local disaster scene for strategic rescue path planning and immediate rescue operations for affected individuals. Unmanned aerial vehicle (UAV) networks are widely utilized for rapid data collection in the aftermath of disasters due to their flexibility and maneuverability, assisting in rescue decision-making. However, some disasters, such as seismic events and floods have disrupted the initially structured ground shape information, leading to a disparate distribution of data collected by various UAV groups. This exposes traditional semantic segmentation models susceptible to shortcut bias, posing challenges in adapting to semantic segmentation tasks in disaster scenarios. Thus, this paper proposes a bias-compensation augmentation learning based semantic segmentation framework, which substantially enhances the extraction capability of semantic information. Initially, we exploit an artificial augmentation neural network for bias-awareness to determine the relative bias values of the collected image data. Subsequently, considering the limited computing power resources in UAV networks, we present a bias compensation computation offloading strategy to achieve a relatively balanced distribution of semantic information across UAV nodes, optimizing the trade-off between network scheduling efficiency and model accuracy. We conduct reconstruction validation on the FloodNet dataset, and a plethora of experimental results demonstrate that, compared to traditional methods, this approach greatly improves the accuracy of pixel-level semantic segmentation by over 86.5%. Moreover, the average combined processing time is also reduced by over 50%, enhancing the utilization efficiency of limited computational resources.

Autonomous robotics play a central role in smart logistics where robots can replace or aid humans in all kinds of tasks, such as items picking, moving, and storing. In this paper, we investigate the problem of task scheduling in automated warehouses with heterogeneous autonomous robotic (HAR) systems. We formulate a long-term non-convex queueing control optimization problem to minimize the queue length of tasks to be processed in the warehouse. Traditional task scheduling solutions based on optimization approaches are inefficient in handling the stochastic nature of the goods/tasks flow and a large number of robots in the system due to their computational cost. We propose a deep reinforcement learning (DRL) based task scheduling algorithm that employs the proximal policy optimization (PPO) method to find an optimal task scheduling policy. Due to the heterogeneity of the system, we propose a proximal weighted federated learning-based algorithm for implementing a decentralized PPO algorithm that improves the performance of the distributed PPO agents that are deployed in the workstations at the geographically distributed warehouses. The simulation results demonstrate the performance improvement of our proposed algorithm compared to the existing methods. Note to Practitioners—Task scheduling for robotic swarms in smart warehouses is substantial for e-commerce. State-of-the-art solutions have focused on efficient task scheduling for homogeneous robotic systems using machine learning techniques implemented in the warehouse management systems (WMS). However, task scheduling for a heterogeneous autonomous robotic (HAR) system has not fully been investigated so far. This article provides a comprehensive task scheduling algorithm for HAR systems that leverages innovative deep reinforcement learning and federated learning techniques. The proposed algorithm can be deployed in the geographically distributed warehouses of an e-commerce company and easily integrated into the WMS to optimally control the operation of the HAR system with stochastic goods/tasks flows in the smart warehousing.

In state-of-the-art autonomous vehicles, data from the vehicle's sensors is often processed using fast and expensive onboard hardware. Such an onboard processing scheme quickly drains the vehicle's battery and consumes computing resources. Recent research proposed to offload parts of processing tasks onto cloud. However, offloading tasks to the cloud is challenging because of the low latency needed for reliable and safe autonomous driving decisions. To address this issue, we propose an Age of Processing (AoP)-aware offloading mechanism for autonomous vehicles. First, we develop a collaboration space of edge clouds to process data closely as possible to the vehicles. Second, we reveal a new communication planning model that allows the vehicle to find suitable open radio units available in route to offload tasks to edge clouds and reduce variation in offloading delay. Third, we formulate an optimization problem that minimizes AoP, i.e., elapsed time from generating tasks and getting computation results. Our AoP-based approach allows a status update to be available to the vehicle after computation. To solve the formulated non-convex problem, we apply dual decomposition and design an AoP-aware algorithm to compute the solution in near real-time. The results demonstrate that our approach meets computation deadlines while minimizing AoP.

5G network provides high-rate, ultra-low latency, and high-reliability connections in support of wireless mobile robots with increased agility for factory automation. In this paper, we address the problem of swarm robotics control for mission-critical robotic applications in an automated grid-based warehouse scenario. Our goal is to maximize long-term energy efficiency while meeting the energy consumption constraint of the robots and the ultra-reliable and low latency communication (URLLC) requirements between the central controller and the swarm robotics. The problem of swarm robotics control in the URLLC regime is formulated as a nonconvex optimization problem since the achievable rate and decoding error probability with short block-length are neither convex nor concave in bandwidth and transmit power. We propose a deep reinforcement learning (DRL) based approach that employs the deep deterministic policy gradient (DDPG) method and convolutional neural network (CNN) to achieve a stationary optimal control policy that consists of a number of continuous and discrete actions. Numerical results show that our proposed multi-agent DDPG algorithm outperforms the baselines in terms of decoding error probability and energy efficiency.

Restoring traffic in multi-layer multi-domain networks (MLMD) can be inefficient and expensive due to the reconfiguration of both intra-domain and inter-domain paths under limited resources and information sharing. This often results in traffic loss and resource over-provisioning within the MLMD, leading to sub-optimal restoration throughput and high costs. In this study, we harness FlexEthernet (FlexE) on inter-domain links to maximize the restoration throughput at minimum cost. FlexE link aggregation is an effective technique to deal with the costly impact of alternative domain rerouting that allows diverting traffic over aggregated links parallel to the failed ones, without disrupting the intra-domain connections. Additionally, FlexE helps increase network reutilization by leveraging time division multiplexing (TDM) to flexibly shift affected traffic to underutilized aggregated links. However, scheduling traffic migration in FlexE is a challenging issue that has not been fully investigated in the literature. In this paper, we initially formulate the FlexE-based traffic restoration problem as a mixed integer non-linear program (MINLP) and then introduce an approximation algorithm to efficiently solve this problem in polynomial time. Furthermore, we propose a supervised learning approach to predict the optimal restoration policy for large-size instances. Experimental results show that our solution restores up to 14% more traffic than a state-of-the-art approach.

The computing power network (CPN) has emerged as a promising networking paradigm in recent times. Since the characteristics of high bandwidth, low delay and high reliable communication, optical networks have been identified as potential frameworks for establishing the CPN infrastructure across metropolitan areas. In CPN of metropolitan areas, owing to the low delay demands of computing power requests, the traffic of computing power requests is more likely to be burst than others. The burst traffic leads to the exponential increase of the traffic loads instantly, which leads to soft failure in the form of overloading and breaks the tradeoff between resource utilization and load balance, which all decline the survivability severely. To solve the problems above, this article proposes an architecture named metro optical computing power network (MO-CPN) to achieve collaborative scheduling in MO-CPN. And proposed a survivable computing power scheduling scheme during burst traffic. Where a multi visual gate recurrent unit (MV-GRU) neural network based on error feedback is constructed to achieve high-precision of burst traffic prediction. According to the burst traffic prediction, a protection threshold to avoid the overloading of nodes is set. And aiming at multi-objectives of low delay and load balancing, the computing power, spectrum resources, burst traffic and protection threshold are used as constraints in the scheduling scheme. The experimental results reveal that our approach can significantly enhance the survivability during burst traffic and improve the utilization of resources. The proposed scheme can also lower the blocking probability and average processing delay, which has strong robustness and reliability.

The Internet of Things (IoT) is an impending area with applications in numerous fields. The number of IoT devices has seen exponential growth, increasing apprehensions around security. Cyberattacks are of rising concern because of the expanded attack surface of threats that have plagued networks. Adding to that are insecure practices among users who may not know to protect their IoT devices. Therefore, IoT security has become fundamental, especially as IoT devices carry sensitive data. This paper provides a proof of concept of an Intelligent Intrusion Detection System for IoT. We centered our work on a privacy-preserving approach offering a Federated Learning (FL) based solution for intrusions recognition combining network and energy data. Our model has achieved high accuracy while preserving a short running time in multiple FL rounds.

Open Radio Access Network (O-RAN) aims to achieve an open and intelligent RAN architecture that provides greater flexibility, scalability, and network optimization. Machine learning (ML) technologies can play a crucial role in achieving these goals by enabling intelligent decision-making, automated optimization, and proactive maintenance. In this paper, we propose an ML pipeline optimization for energy-efficient deployment of ML-based closed-loop controls (CLC) in 5G O-RAN. Specifically, we propose two ML-based CLCs for resource prediction and network slicing in Non-Realtime RIC and Near-Realtime RIC. We also propose an energy-efficient ML pipeline for dynamically deploying these two CLCs in the O-RAN architecture. Our numerical results demonstrate the effectiveness of our proposed ML pipeline deployment compared to fixed centralized and distributed deployment.

Traditional Fronthaul architectures are not efficiently provisioned for 5G due to their lack of Fronthaul slicing supports or/and inability to satisfy the very strict latency and high bandwidth requirements defined for the enhanced common public radio interface (eCPRI). In this paper, we propose a new Fronthaul architecture that leverages Flex-Ethernet (FlexE) to guarantee 5G Fronthaul quality of service (QoS) requirements without over-provisioning the Fronthaul resources. By separating the MAC and PHY layers through a time division multiplexing (TDM) shim, FlexE allows for efficient aggregation of huge traffic volume and the design of low-latency hard network slicing architectures. Unfortunately, no standard has been defined for eCPRI transmission support over FlexE. Therefore, we propose a new protocol stack in which FlexE clients are efficiently allocated to carry eCPRI data of different 5G slices. We then formulate the FlexE-based Fronthaul provisioning optimization problem as an integer linear program (ILP) model. Simulation results show the proposed solution saves 82% of CAPEX and 94% of OPEX compared to a Fronthaul baseline.

For digital inclusion, Internet quality in Low-Density and Rural Areas (LDRAs) should be enhanced to satisfy QoS requirements of various services and applications. Due to the high operating costs of fiber optic deployment in LDRAs, 5G Fixed Wireless Access (5G FWA) is becoming a more attractive solution. Furthermore, 5G services require edge cloud deployment for proximity computation, which increases both required network and energy resources. Therefore, we propose closed-loops assisted by Digital Twin (DT) for energy-efficient Open RAN-based FWA provisioning in LDRAs. We consider a 5G FWA and edge cloud system as Physical Twin (PT) and design a closed-loop that distributes radio resources to edge cloud instances that manage network slices for scheduling purposes. We propose another closed-loop for intra-slice resource allocation to LDRAs. We develop an energy model and join radio resource allocation with the energy model. Then, we design reinforcement learning and optimization approaches to maximize delay requirement satisfaction while minimizing energy cost. Finally, we present DT replicating PT by incorporating solution experiences into future states. The results show that our approach uses energy resources efficiently while satisfying delay requirements of slices.

In this paper, we consider a UAV-assisted cloud robotic network in which a set of robots is deployed to perform specific missions, e.g., surveillance and rescue, in an area where the communication condition is unfavorable for the robots. Data collected by a robot can be either offloaded to a MEC server or to a remote cloud through the UAVs or to a nearby robot for computation. We formulate this offloading problem as a combinatorial nonconvex problem. A joint scheme for offloading decision-making, robot-UAV association, and computational resource allocation is proposed using KKT conditions, Lagrangian dual decomposition, and the Proximal Policy Optimization method to obtain the solution to the formulated problem. The simulation results show our proposed algorithm achieves a solution close to the optimal solution and outperforms the baselines.

5G intends to use network slicing to support multiple vertical industries such as the power grid. 5G network slicing can provide different levels of physical resources and virtual resources for various applications/services in vertical domains to meet their diversified communication requirements. These heterogeneous Service Level Agreements (SLAs) make the network highly dynamic in nature and challenging to operate and manage efficiently. In this paper, we formulate the SLA-based closed-loop automation network slicing management problem for 5G smart grid services in Open Radio Access Network (O-RAN). The resource scheduling problem is non-convex combinatorial while the resource reservation is a long-term mean-square-error minimization which is difficult to solve. We propose a collaborative game theory and deep learning solution that overcomes the complexity difficulty of the formulated problems. The proposed network slicing mechanism comprises three closed-loop control: closed-loop 1 resource request at the service layer, closed-loop 2 resource scheduling at the radio access layer, and closed-loop 3 resource reservation at the network layer. Simulation results show that the proposed slicing framework is more efficient than the baselines regarding fairness and network throughput.

Recently, IoT, SDN and NFV have emerged as significant technological enablers for telemedicine. Because of specific characteristics of telemedicine services, reliability is one of the critical elements to guarantee the quality of services. To maintain high availability of services, existing backup solutions focus on resources constraints where backup instances are placed at the same node (vertical backup) or distributively deployed at different nodes (horizontal backup). While they put more effort to satisfy resource requirements, routing issues are often neglected such as end-to-end latency, multi-path routing, and synchronization in a multi-path scenario. Such aspects are key requirements to deploy high reliability telemedicine services. Therefore, we investigate the dynamic backup mechanism for a telemedicine system. We aim to optimize the reliability of telemedicine service function chains (TSFCs) where a joint horizontal/vertical backup (JHVB) optimization problem is first formulated. Since JHVB is a combinatorial optimization problem, which is NP-Hard, we then solve this problem in both offline and online fashions using Block Successive Upper Bound Minimization (BSUM) and Multi-Armed Bandit (MAB) frameworks. We compare our methods to the benchmarks via intensive simulations based on the Nano Datacenter solution that is used for enabling telemedicine services. The results demonstrates an outstanding performance in terms of failure awareness and service reliability.

The orchestrator coordinates multiple layers, such as IP, OTN, and DWDM, in a Multi-Layer Network (MLN), in which each layer is a domain. Accurately updating the orchestrator's network view is a key challenge in MLNs. Due to the hierarchical structure of MLNs, a single failure in an underlying layer can propagate to the upper layers, hence generating many alarms in each of these layers. These alarms may send incorrect updates of the root cause of the failure, which results in incorrect decisions of the orchestrator for routing and resource allocation. Therefore, setting up the order of updating the orchestrator by different layers is crucial in MLN to avoid confusion in the orchestrator's network view. This task is challenging, due to the flexible mapping of links between different layers, and also to the failure propagation time from the underlying layers to the upper layers. In this paper, we propose a method to update the orchestrator to ensure that the root cause is reported correctly, taking into account the dependency among different layers and failure propagation time. Moreover, to compute the optimal frequency of sending update messages from layers to the orchestrator. Our proposed method can be implemented in the Topology Server (TS) of the orchestrator. We formulate an integer nonlinear optimization problem for updating the orchestrator and then propose an algorithm to approximate the optimal failure probability for updating the orchestrator. Simulation results show that our algorithm can obtain a near-optimal which is, on average 11.3% different from the global minimum.

The Internet of Things (IoT) plays a critical role in enabling the seamless integration of disparate devices. Future IoT will rapidly expand its coverage to offer future worldwide omnipresent applications and services by merging communications in diverse spatial domains to build the space–air–ground–ocean-integrated network (SAGOI-Net). SAGOI-Net will include a significant number of battery-powered network nodes, such as satellites, unmanned vehicles, and underwater devices, due to the extremely vast geographic reach and dynamics in free space. Given the battery limitations, high-energy-efficiency communications and networking will be critical to the future system. Green SAGOI-Net seeks to not only bring ubiquitous connectivity to every corner of the globe but also to deliver more data with the same amount of energy.

Unmanned aerial vehicles (UAVs) are projected to be utilized in a variety of unexpected applications, including agriculture, firefighting, emergency response, intelligent transportation, and so on. Wireless communication is one of the primary facilitators in bringing UAVs into a new phase in such applications. To realize the vision in fifth-generation (5G) networks, we propose a 5G-integration of the flexible multi-UAV system and the Open Radio Access Network (O-RAN) architecture, named U-ORAN. Although different studies have been proposed to optimize the UAV trajectory and resource allocation in the radio access network (RAN), our work is the first study to investigate the benefits of adopting UAVs in the O-RAN architecture. In U-ORAN, we consider a flying base station system and propose a joint optimization problem of multi-UAV trajectory and offloading tasks (UTOT) in which UTOT can optimize the routes from users to the core network as well as resource allocation to process offloading tasks. We decompose UTOT into two sub-problems and provide learning solutions based on the multi-agent reinforcement learning and online learning methodologies, both of which are well supported by the O-RAN architecture. Our intensive numerical simulations show that the proposed approaches outperform in a variety of settings and validation scenarios.

The fifth-generation (5G) wireless network provides high-rate, ultra-low latency, and high-reliability connections that can meet the Industrial Internet of Things (IIoT) requirements in factory automation, especially for robot motion control. In this paper, we address 5G service provisioning in an automated warehouse scenario, where swarm robotics is controlled by an industrial controller that provides routing and job instructions over the 5G network. Leveraging the coordinated multipoint (CoMP), we formulate a time-varying joint CoMP clustering and 5G ultra-reliable low-latency communication (URLLC) beamforming design problem to control the robots that move around the automated warehouse for goods storage with the planned reference tracks. Traditional iterative optimization approaches are impractical in such a dynamic wireless environment due to high computational time. We propose a game-theoretic CoMP clustering algorithm combined with the Proximal Policy Optimization method to obtain a stationary solution closed to that of the exhaustive search algorithm considered as the global optimal solution.

Flying and ground-based cars require various services such as autonomous driving, remote piloting, infotainment, and remote diagnosis. Each service requires specific Quality of Service (QoS) and network features. Therefore, network slicing can be a solution to fulfill the requirements of various services. Some services, such as infotainment, may have similar requirements to serve flying and ground-based cars. Therefore, some slices can serve both kinds of cars. However, when network slice resource sharing is too aggressive, slices can not meet QoS requirements, where resource under-provisioning causes the violation of QoS, and resource over-provisioning causes resources under-utilization. We propose two closed loops for managing RAN slice resources for cars to address these challenges. First, we present an auction mechanism for allocating Resource Block (RB) to the tenants who provide services to the cars using slices. Second, we design one closed loop that maps slices and services of tenants to Open Distributed Units (vO-DUs) and assigns RB to vO-DUs for management purposes. Third, we design another closed loop for intra-slices RB scheduling to serve cars. Fourth, we present a reward function that interconnects these two closed loops to satisfy the time-varying demands of cars at each slice while meeting QoS requirements in terms of delay. Finally, we design distributed deep reinforcement learning approach to maximize the formulated reward function. The simulation results show that our approach satisfies more than 90% vODUs resource constraints and network slice requirements.

Offloading computation-intensive tasks to edge clouds has become an efficient way to support resource constraint edge devices. However, task offloading delay is an issue largely due to the networks with limited capacities between edge clouds and edge devices. In this paper, we consider task offloading in Open Radio Access Network (O-RAN), which is a new 5G RAN architecture allowing Open Central Unit (O-CU) to be co-located with Open Distributed Unit (DU) at the edge cloud for low-latency services. O-RAN relies on fronthaul network to connect O-RAN Radio Units (O-RUs) and edge clouds that host O-DUs. Consequently, tasks are offloaded onto the edge clouds via wireless and fronthaul networks, which requires routing. Since edge clouds do not have the same available computation resources and tasks’ computation deadlines are different, we need a task distribution approach to multiple edge clouds. Prior work has never addressed this joint problem of task offloading, fronthaul routing, and edge computing. To this end, using segment routing, O-RAN intelligent controllers, and multiple edge clouds, we formulate an optimization problem to minimize offloading, fronthaul routing, and computation delays in O-RAN. To determine the solution of this NP-hard problem, we use Deep Q-Learning assisted by federated learning with a reward function that reduces the Cost of Delay (CoD). The simulation results show that our solution maximizes the reward in minimizing CoD.

5G and beyond network will support vertical industry applications, and the resource requirements of each service vary widely. The introduction of network slices provides great flexibility to the network, which can realize the differentiated customization requirements of service. However, while determining how to intelligently orchestrate the network slices is an important challenge, current solutions rarely treat multiple customized requirements of delay, bandwidth, load balancing, and slice isolation. In this article, network slice orchestration is considered from the perspective of slice isolation and cloud-edge collaboration. First, differentiated isolation level requirements are restricted to constraints, the customized isolation is realized. Second, bandwidth is saved and network latency is reduced via the collaboration of cloud and edge data centers. In addition, exclusive orchestration optimization objectives that match various service needs are proposed to distinguish the specific requirements of different slices. Finally, two deep reinforcement learning-based algorithms are proposed. The experimental results demonstrate that the proposed algorithms can optimize the objectives while ensuring differentiated isolation levels. For typical slices, the proposed algorithms respectively reduce bandwidth consumption by about 29% and 64%, reduce slice delay by about 14% and 70%, and optimize load balancing by about 17% and 23%.

In this paper, we investigate the problem of task scheduling in automated warehouses with hetero-geneous autonomous robotic systems. We formulate the task scheduling for a heterogeneous autonomous robots (HAR) system in each warehouse as a queueing control optimization problem in which we aim to minimize the queue length of tasks that are waiting to be processed. We propose a deep reinforcement learning (DRL) based approach that employs the proximal policy optimization (PPO) to achieve an optimal task scheduling policy. We then propose a federated learning based algorithm to improve the performance of the PPO agents. The simulation results fully demonstrate the performance improvement of our proposed algorithm in terms of average queue length compared to the distributed learning algorithm.

Unmanned aerial vehicles (UAVs) are increasingly deployed as flying base stations to serve various applications, such as smart agriculture, emergency healthcare system, smart transportation, etc, thanks to their advantages of flexible movement, strong wireless communication, and heavy payload capability. To facilitate the deployment of the fifth-generation (5G) networks, the Open Radio Access Network (O-RAN) has presented a distributed architecture for terrestrial and non-terrestrial networks. Unfortunately, O-RAN architecture for wireless sensor networks is still in development. In this paper, we investigate a 5G integration of multi-flying base stations in a wireless sensor network using O-RAN. Specifically, we formulate a joint optimization problem of UAV trajectory and resource allocations to process sensing data, named UTRA. We use decomposition to address it based on two solvable sub-problems and provide learning methodologies solve UTRA based on the multi-agent reinforcement learning and online learning methods, both of which are well supported by the O-RAN architecture. Our extensive numerical simulations show that our proposed approaches are efficient in a variety of settings and validation scenarios.

Open Radio Access Network (O-RAN) is an innovative RAN architecture designed to revolutionize 5G-and-beyond mobile networks. O-RAN virtualizes the fronthaul network functions into Open Centralized Unit (O-CU), Open Distributed Unit (O-DU) and Open Radio Unit (O-RU). Unfortunately, there is no standard data communication mechanism to disaggregate Quality of Service (QoS) flow traffic into multiple routes to access O-DUs to leverage the distributed computing capability. Furthermore, there is no centralized scheduler to coordinate processors that are processing O-DU functions efficiently to meet fifth generation (5G) QoS services. Therefore, O-RAN performance is still questionable. This paper investigates an optimized solution for joint Routing and Packet Scheduling (RPS) which is implemented in the O-DU pool to replace individual O-DUs. We formulate two joint RPS problems to coordinate multiple routes and multiple parallel processors in the centralized O-DU pool to accommodate the Ultra-Reliable Low Latency Communications (URLLC) and enhanced Mobile Broadband (eMBB) services. We propose a greedy algorithm and a Min-Max algorithm to approximate the optimal result. Numerical results show that our proposed solution improves significantly system processing delay compared with a scheme of individual O-DUs which are selfishly maximized.

During the last few years, research in historical document restoration and understanding (HDRU) has gained increasing popularity. One major problem facing HDRU is the presence of degradation, which renders historical documents unreadable. Although promising results have been obtained, these methods lack the ability to generalize across different datasets. Also, multiple pre-processing and post-processing steps are used, which add more computational complexity and make inference unpractical in real-life settings. Deep Learning (DL) has been successfully used to solve various supervised learning problems in computer vision, where labeled datasets are readily available. However, in HDRU, large annotated historical document datasets are not available. In this paper, we propose an efficient multitask learning (MTL) approach that is based on jointly training self-supervised and supervised learning modules. In the self-supervised learning module, we define two tasks that can be trained with unlabeled data. The first task consists of denoising, and the second task is to learn handwritten characteristics (text orientation). In the supervised learning module, a small subset of labeled data is used to perform text extraction or binarization. All the tasks are formulated as a multi-objective Frank-Wolfe-based optimization problem. We show that convergence to a Pareto optimal solution of jointly training multiple tasks together improves the overall invariance and accuracy of the model. DIBCO 2010-2017 datasets were used for training and DIBCO 2018 for testing. We achieved state-of-the-art results with an F-Score measure of 91.21.

The fifth-generation (5G) wireless net-work provides high-rate, ultra-low latency, and high-reliability connections that can meet the industrial IoT requirements in factory automation especially for swarm robotics communication. In this paper, we address 5G service provisioning in an automated warehouse scenario where swarm robotics is controlled by an industrial controller that provides routing and job instructions over the 5G network. Leveraging the co-ordinated multipoint (CoMP), we formulate a joint CoMP clustering and 5G ultra-reliable low-latency communication (URLLC) beamforming design problem to control the robots that move around the automated warehouse for goods storage with the planed reference tracks. Traditional iterative optimization approaches are impractical in such dynamic wireless environments due to high computational time. We propose a game-theoretic CoMP clustering algorithm combined with the Proximal Policy Optimization method to obtain a stationary solution closed to that of the exhaustive search algorithm considered as the global optimal solution.

In a multiple-input multiple-output frequency-division duplexing (MIMO-FDD) system, the user equipment (UE) sends the downlink channel state information (CSI) to the base station to report link status. Due to the complexity of MIMO systems, the overhead incurred in sending this information negatively affects the system bandwidth. Although this problem has been widely considered in the literature, prior work generally assumes an ideal feedback channel. In this paper, we introduce PRVNet, a neural network architecture inspired by variational autoencoders (VAE) to compress the CSI matrix before sending it back to the base station under noisy channel conditions. Moreover, we propose a customized loss function that best suits the special characteristics of the problem being addressed. We also introduce an additional regularization hyperparameter for the learning objective, which is crucial for achieving competitive performance. In addition, we provide an efficient way to tune this hyperparameter using KL-annealing. Experimental results show the proposed model outperforms the benchmark models including two deep learning-based models in a noise-free feedback channel assumption. In addition, the proposed model achieves an outstanding performance under different noise levels for additive white Gaussian noise feedback channels.

The Open Radio Access Network (O-RAN) architecture offers flexible association of distributed network elements, i.e., O-RAN elements, for establishing on-demand radio access network (RAN) stacks [1]. In such as distributed architecture, content caching may help relieve network traffic congestion and end-to-end latency. Leveraging the strengths of both O-RAN architecture and content caching is thus essential for reducing network traffic while meeting user’s quality of experience (QoE) requirements. However, integrating content caching into O-RAN architecture is not straightforward as the content placement policy and the O-RAN element association are tightly coupled. In this paper, we investigate the problem of joint O-RAN element association and content placement in the O-RAN architecture, aiming at minimizing the average long-term traffic incurred in the xHaul transport networks. To this end, a line-search based algorithm is proposed to find efficient O-RAN element association solution; meanwhile, a deep reinforcement learning (DRL) based algorithm is proposed to obtain an efficient content placement policy while being able to deal with large action space issue. The efficacy of our proposed framework is confirmed through extensive numerical results.

Recently, wireless service provisioning via Unmanned Aerial Vehicles (UAVs) has emerged in 5G and beyond mobile networks. Due to the limited capacity of UAV batteries, tethered UAVs (TUAVs), which are powered from ground, are increasingly deployed in worldwide projects. However, the deployment of TUAVs in mobile networks requires high spectral efficiency, particularly in dense areas. Non-orthogonal Multiple Access (NOMA), serving users with strong channels and weak channels in the same Resource Blocks (RBs), helps overcome this issue. Due to the dynamic and massive deployment of TUAVs, inter-cell interference becomes critical. To alleviate the submerging of signals between TUAVs, we introduce a new parameter named Channel Gain Plus Interference (CGPI), reputing the interference as channel characteristics. Then, we formulate the joint optimization of power, altitude and user association. To solve this high-complexity problem, we design an algorithm, called NOMA SIC-Aware TUAV Base Station Control (NSATC) based on Multi-Agent Deep Deterministic Policy Gradient (MADDPG). The experiment shows our proposed algorithm presents a performance enhancement between 18.8% and 121.77% of throughput and 23.76% and 51.62% of serving users than the greedy algorithm.

Recently, the exponential growth of the Internet-of-Things (IoT) services with heterogeneous requirements becomes a burden to the traditional cloud/data center platform. Edge computing is an emerging solution to gain business value of IoT services where real-time demands become satisfied by moving computing resources close to data sources. Nevertheless, edge resources are still limited to be able to fulfill all demands at the same time. Among new approaches, resource sharing between edge/cloud service providers has been considered as a promising mechanism to address resource scarcity and pursue cost reduction. In this article, we propose an allocation and sharing model in the edge cloud network where providers team up to efficiently utilize resources, named the share-to-run IoT services (SRIS). In particular, we formulate a resource allocation and sharing optimization model to implement IoT services of multiple edge/cloud providers that can maximize the providers’ utility while satisfying service constraints. We relax SRIS into a tractable form that can be solved efficiently using well-known distributed convex frameworks, such as the dual decomposition and alternating direction method of multipliers. Finally, we evaluate our methods by providing several simulation cases, in which our proposed mechanisms show outstanding outcomes by obtaining a faster convergence, increasing by 6.9% of utilization, and 16% of acceptance rate compared to the nonoptimal approach.

This paper proposes a novel Nonlinear Orthogonal NMF model with Graph-based Total Variation regularization (GTV) for Multispectral document images decomposition. In this model, a GTV regularization is incorporated to preserve the intrinsic geometrical structure of document content lost by the vectorization of spectral images. A spatial orthogonality constraint over the Stiefel manifold is imposed to improve the sparsity of the solution and ensure its uniqueness. The kernel trick is involved to account for the non-linear correlation inherent to spectral data. We devised an efficient algorithm to solve the formulated problem using the Alternating Direction Method of Multipliers (ADMM). The experimental results on real-world data show that the proposed model achieves better decomposition performance than recent competitive methods and outperforms some traditional state-of-the-art methods.

The Industrial Internet of Things (IIoT) is one of the important applications under the 5G massive machine type of communication (mMTC) scenario. To ensure the high reliability of IIoT services, it is necessary to apply an efficient resource allocation method under the dynamic and complex environment. In view of the absence of energy-efficient resource management architecture for the entire network, this article proposes an intelligent-driven green resource allocation mechanism for the IIoT under 5G heterogeneous networks. First, an intelligent end-to-end self-organizing resource allocation framework for IIoT service is given. Next, an energy-efficient resource allocation model within the framework is proposed. It is then solved by an intelligent mechanism with the asynchronous advantage actor critic driven deep reinforcement learning algorithm. Through the comparison analysis of different methods and rewards under IIoT scenarios with proper parameters setting, the proposed method can achieve better performance than other traditional deep learning (DL) methods and maintain service quality above accepted levels as well.

Nonnegative matrix factorization (NMF) is one of the best-known multivariate data analysis techniques. The NMF uniqueness and its rank selection are two major open problems in this field. The solutions uniqueness issue can be addressed by imposing the orthogonality condition on NMF. This constraint yields sparser part-based representations and improved performance in clustering and source separation tasks. However, existing orthogonal NMF algorithms rely mainly on non-probabilistic frameworks that ignore the noise inherent in real-life data and lack variable uncertainties. Thus, in this work, we investigate a new probabilistic formulation of orthogonal NMF (ONMF). In the proposed model, we impose the orthogonality through a directional prior distribution defined on the Stiefel manifold called von Mises-Fisher distribution. This manifold consists of a set of directions that comply with the orthogonality condition that arises in many applications. Moreover, our model involves an automatic relevance determination (ARD) prior to address the model order selection issue. We devised an efficient variational Bayesian inference algorithm to solve the proposed ONMF model, which allows fast processing of large datasets. We evaluated the proposed model, called VBONMF, on the task of blind decomposition of real-world multispectral images of ancient documents. The numerical experiments demonstrate its efficiency and competitiveness compared to the state-of-the-art approaches.

Today, vehicles use smart sensors to collect data from the road environment. This data is often processed onboard of the vehicles, using expensive hardware. Such onboard processing increases the vehicle’s cost, quickly drains its battery, and exhausts its computing resources. Therefore, offloading tasks onto the cloud is required. Still, data offloading is challenging due to low latency requirements for safe and reliable vehicle driving decisions. Moreover, age of processing was not considered in prior research dealing with low-latency offloading for autonomous vehicles. This paper proposes an age of processing-based offloading approach for autonomous vehicles using unsupervised machine learning, Multi-Radio Access Technologies (multi-RATs), and Edge Computing in Open Radio Access Network (O-RAN). We design a collaboration space of edge clouds to process data in proximity to autonomous vehicles. To reduce the variation in offloading delay, we propose a new communication planning approach that enables the vehicle to optimally preselect the available RATs such as Wi-Fi, LTE, or 5G to offload tasks to edge clouds when its local resources are insufficient. We formulate an optimization problem for age-based offloading that minimizes elapsed time from generating tasks and receiving computation output. To handle this non-convex problem, we develop a surrogate problem. Then, we use the Lagrangian method to transform the surrogate problem to unconstrained optimization problem and apply the dual decomposition method. The simulation results show that our approach significantly minimizes the age of processing in data offloading with 90.34% improvement over similar method.

With the capability of decoupling the control plane and the data plane of networks, Software-Defined Network (SDN) enables flexible and efficient implementations in networks. In addition, Network Function Virtualization (NFV) with Virtual Network Function (VNF) service chain capabilities provides high-performance networks with greater scalability, elasticity, and adaptability. Such an elastic deployment of service chains results in different Service Level Agreements (SLA) and resource requirements on the control plane. In this work, we illustrate the impact of service chains on the control plane and formulate the dynamic controller/switch mapping (DCSM) problem in NFV networks in order to reduce the operational cost. We address the combinatorial optimization problem, DCSM, by designing a novel mechanism to relax DCSM into a tractable problem based on the Penalty Successive Upper Bound Minimization (PSUM) method. In doing so, we conduct several simulation scenarios to evaluate the performance. The experimental results show that our proposed algorithms can achieve a near-optimal result and reduce the operational cost up to 31.7% and 28.3% compared to K-Mean and the matching game-based approaches, respectively.

Energy consumption is becoming a key issue in the research of future network. In practice, network traffic has a periodic time distribution that occurs most often at a low level. This feature provides the possibility of achieving network energy savings through topology switching. By considering the deficiencies in existing studies, such as the low adaptability between network working topology and traffic load, the abnormal topology switching caused by abnormal and unbalanced traffic, and the low reliability of energy-saving topology, this paper proposes an energy-efficient routing method for software-defined networks based on topology switching and reliability. The method involves two parts: a topology-switching method and a failure recovery method. The former adapts the network working topology to the network traffic demands through dynamic topology switching to decrease the network energy consumption. The latter adopts an active strategy for fast fault recovery to ensure network reliability in the energy-efficient topology. Two network typologies and their traffic data are used to experimentally verify the method. The results show that, compared with the static topology switching method TLS, the energy saving of the proposed method can be improved at most 2.07 times and 4.63 times in two typical typologies, respectively, while ensuring network reliability.

Today, Machine Learning (ML) techniques are increasingly used to detect abnormal behaviours of industrial applications. Since many of these applications are moving to the cloud environments, classical ML approaches are facing new challenges in accurately identifying abnormal behaviours due to the highly dynamic and heterogeneous nature of the cloud. In this paper, we propose a novel framework, DistAppGaurd, for profiling simultaneously the behaviour of all microservice components of a distributed application in the cloud. The framework can therefore, detect complex attacks that are not observable by monitoring a single process or a single microservice. DistAppGaurd utilizes the system calls executed by all the processes of an application to build a graph consisting of data exchanges among different application entities (e.g., processes and files) representing the behaviour of the application. This representation is then used by our novel miroservice-aware Autoencoder model to perform anomaly detection at runtime. The efficiency and feasibility of our approach is shown by implementing several different real-world attacks, which yields high detection rates (94%-97%) at 0.01% false alarm rate.

In IoT applications, the sensors usually have limited bandwidth and power resources. Therefore, the sensed data should be mapped to a low-bit representation by means of compression and quantization before being transmitted to a central node, called the fusion center (FC). At the FC, a global decision is inferred from this data. In many cases, this data is intended for machine consumption, not for human perception. However, the compression techniques are mainly designed for reconstruction fidelity. The accuracy of the inferred decision at the FC is less considered. In this work, we present an end-to-end framework for learning a 1-bit representation of correlated-sensors data. We also propose a novel loss function and a three-stage training algorithm for learning discriminative binary features at each sensor. Extensive experiments show the proposed framework achieves high compression ratios with a marginal loss in the inferred decision accuracy. Comparatively, the obtained results outperform other benchmark models in the literature.

Unmanned aerial vehicle (UAV)-assisted communications have several promising advantages, such as the ability to facilitate on-demand deployment, high flexibility in network reconfiguration, and high chance of having line-of-sight (LoS) communication links. In this paper, we aim to optimize the UAV control for maximizing the UAV's energy efficiency, in which both aerodynamic energy and communication energy are considered while ensuring the communication requirements for each ground terminal (GT) and backhaul link between the UAV and the terrestrial base station (BS). The mobility of the UAV and GTs lead to time-varying channel conditions that make the environment dynamic. We formulate a nonconvex optimization for controlling the UAV considering the practical angle-dependent Rician fading channels between the UAV and GTs, and between the UAV and the terrestrial BS. Traditional optimization approaches are not able to handle the dynamic environment and high complexity of the problem in real-time. We propose to use the Trust Region Policy Optimization (TRPO) method that can improve the performance of the UAV compared to the Deep Deterministic Policy Gradient (DDPG) method in such a dynamic environment as in this paper.

In this paper, we investigate the problem of robot swarm control in 5G mission-critical robotic applications, i.e., in an automated grid-based warehouse scenario. Such application requires both the kinematic energy consumption of the robots and the ultra-reliable and low latency communication (URLLC) between the central controller and the robot swarm to be jointly optimized in real-time. The problem is formulated as a nonconvex optimization problem since the achievable rate and decoding error probability with short block-length are neither convex nor concave in bandwidth and transmit power. We propose a deep reinforcement learning (DRL) based approach that employs the deep deterministic policy gradient (DDPG) method and convolutional neural network (CNN) to achieve a stationary optimal control policy that consists of a number of continuous and discrete actions. Numerical results show that our proposed multi-agent DDPG algorithm achieves a performance close to the optimal baseline and outperforms the single-agent DDPG in terms of decoding error probability and energy efficiency.

Open Radio Access Network (O-RAN) is an innovative RAN architecture designed to revolutionize 5G and beyond mobile networks. O-RAN virtualizes the fronthaul network functions into O-RAN Centralized Unit (O-CU), O-RAN Distributed Unit (O-DU) and O-RAN Radio Unit (O-RU). Unfortunately, no standard data communication mechanism has been defined for the communication between these elements. Therefore, O-DUs may not work efficiently in O-DU pool, limiting the RAN performance. This paper investigates an optimized solution for routing and packet scheduling, allowing multiple O-DU pools to communicate with their O-RUs meeting the requirements of different 5G classes of service. We propose an O-DU pool architecture and formulate the problem of optimal routing and packet scheduling to forward Orthogonal Frequency-Division Multiplexing (OFDM) symbols over the optimal routes and map UDP packet sizes to fragment OFDM symbols. Numerical results show our solution can select the optimal routes and packet sizes to carry requested traffic. Moreover, in the multiple O-DU pools coexisting, we use the Dynamic Programming (DP) algorithm to find out the optimal global solution and a greedy algorithm to approximate the solution in near real-time.

As a nonlinear extension of the standard nonnegative matrix factorization (NMF), kernel-based variants have demonstrated to be more effective for discovering meaningful latent features from raw data. However, many existing kernel methods allow only obtaining the basis matrix in the projected feature space, which prevents its inverse mapping back to the original space as requested in many applications. In this work, we propose a new kernel orthogonal NMF method that does not suffer from the pre-image issue. We incorporate the orthogonality constraint as an optimization problem over the Stiefel manifold to improve the sparsity and the model’s clustering properties. We solve the proposed model with an efficient optimization approach based on the alternating direction method of multipliers (ADMM) scheme and the projected gradients method. We validate our model on the task of blind decomposition of real-world Multispectral (MS) document images. Our experiments demonstrate the competitiveness of our proposed model in comparison to the state-of-the-art techniques.

The problem of selecting the modulation and coding scheme (MCS) that maximizes the system throughput, known as link adaptation, has been investigated extensively, especially for IEEE 802.11 (WiFi) standards. Recently, deep learning has widely been adopted as an efficient solution to this problem. However, in failure cases, predicting a higher-rate MCS can result in a failed transmission. In this case, a retransmission is required, which largely degrades the system throughput. To address this issue, we model the adaptive modulation and coding (AMC) problem as a multi-label multi-class classification problem. The proposed modeling allows more control over what the model predicts in failure cases. We also design a simple, yet powerful, loss function to reduce the number of retransmissions due to higher-rate MCS classification errors. Since wireless channels change significantly due to the surrounding environment, a huge dataset has been generated to cover all possible propagation conditions. However, to reduce training complexity, we train the CNN model using part of the dataset. The effect of different subdataset selection criteria on the classification accuracy is studied. The proposed model adapts the IEEE 802.11ax communications standard in outdoor scenarios. The simulation results show the proposed loss function reduces up to 50% of retransmissions compared to traditional loss functions.

Recently, Flexible Ethernet (FlexE) has emerged as a new transmission technology allowing the flexible utilization of optical transport. This flexibility helps improve network ability against failures. FlexE recovers from a physical link (PHY) failure by migrating traffic to a new PHY. However, this task is costly, especially for critical failures or when the network is under high utilization. In this paper, we investigate the FlexE Traffic Restoration (FTR) problem that aims to maintain high network utilization by the fast recovery of FlexE clients with the minimum cost using the spare capacity in the already deployed PHYs. High network utilization can be obtained by rerouting of FlexE subgroups, moving clients to another subgroup, and shifting the clients’ slots in the same subgroup without traffic disruption. We formulate the FTR optimization problem and solve it in polynomial time using learning theory and approximation. Experiments carried out in a real testbed show the proposed solution recovers 63% more traffic than baseline restoration schemes.

In the burgeoning age of digitalization, the Internet of Things presents a core part of the digital ecosystem. Unfortunately, as the deployment of connected devices is increasing tremendously, so are cyber-attacks. The consequences of cyber-attacks could be devastating as they gain access to sensitive data and even damages critical infrastructures. This urges the development and integration of proactive and intelligent security breach detection mechanisms in different levels of the IoT platforms including the devices themselves. Several empirical observations indicated a change in the energy consumption and network behaviour of compromised devices. Thus, we propose in this paper a machine learning based approach to identify compromised IoT devices using their energy consumption footprint and network traffic. We base our study on real data collected from real experiments using different commercially available IoT devices infected with authentic IoT botnets. Our results show that machine learning algorithms can classify correctly attacks reaching 98.40% precision for Mirai, over 99.91% for Ufonet and respectively 97.63% and 99.93% performance. Overall, our exploratory study is one of the very first of its kind to explore the energy consumption combined with network behavior analysis to detect IoT compromised devices and its outcomes will be a starting point for further research on this topic.

The Industrial Internet of Things (IIoT) has attracted tremendous interest from both industry and academia as it can significantly improve production efficiency and system intelligence. However, with the explosive growth of various types of user equipment (UE) and data flow, IIoT experiences spectrum resource scarcity for wireless applications. In this article, we propose a solution for spectrum resource management for the IIoT network, with the objective of facilitating the limited spectrum sharing between different kinds of UEs. To overcome the challenges of unknown dynamic IIoT environments, a modified deep $Q$ -learning network (MDQN) is developed. Considering the cost effectiveness of IIoT devices, the base station (BS) acts as a single agent and centrally manages the spectrum resources, which can be executed without coordination or exchange between UEs. In this article, we first built a realistic IIoT model and design a simple medium access control (MAC) frame structure to facilitate the environment state observation. Then, a new reward function is designed to drive the learning process, which takes into account the different communication requirements of various types of UEs. In addition, to improve the learning efficiency, we compress the action space and propose a priority experience replay strategy based on decreasing temporal difference (TD) error. Finally, simulation results show that the proposed algorithm can successfully achieve dynamic spectrum resource management in the IIoT network. Compared with other algorithms, it can achieve superior network performance with a faster convergence rate.

The combination of space division multiplexing technology with elastic optical networks allows to overcome the possible capacity crunch in backbone networks and also improves network flexibility by jointly managing spectral and spatial resources. However, against this background implemented by multi-core fibers, the interaction between spatial modes will appear as signal crosstalk, thereby affecting the service’s transmission quality. Spectrum resources without crosstalk are always preferred for the services to guarantee quality of service, possibly resulting in the spectrum fragmentation. Conversely, if resources with crosstalk are selected for services to reduce fragments, it may lead to serious crosstalk on the services already carried in the adjacent cores. To achieve a tradeoff between these two factors, this paper firstly exploits the association rule mining method to quantitatively analyze the potential correlation between them. By executing FP-growth mining algorithm, rules not beneficial to service provisioning will be filtered out. Then, an association rules-based core and spectrum assignment algorithm is presented, considering transmission requirements for different levels of services. Simulation results indicate the presented strategy can decrease the proportion of services affected by crosstalk and also reduce the possibility of fragments generation. Additionally, it can effectively make improvement on the blocking and resource utilization.

This paper addresses the challenge of Multispectral (MS) document image segmentation, which is an essential step for subsequent document image analysis. Most previous studies have focused only on binary (text/non-text) separation. They also rely on handcrafted features and techniques dedicated to conventional images that do not take advantage of MS images' spectral richness. In this work, we reformulate this task as a source separation problem, whereby we target the blind decomposition of entire MS document images via a new orthogonal nonnegative matrix factorization (ONMF). On the one hand, we incorporate orthogonality constraint as a Riemannian optimization on the Stiefel manifold. On the other hand, based on which factor we impose the orthogonality constraint, i.e., either on the endmember matrix, abundance matrix, or both, we propose three ONMF models to investigate this issue and determine which model is more suitable for this study. Minimizing the three models subject to nonnegativity and orthogonality constraints simultaneously is very challenging. Therefore, we extend the alternating direction method of multipliers scheme to solve them. We evaluated our models on synthetic Hyperspectral (HS) images and real-world MS document images. The experimental results confirm the effectiveness of the proposed models and demonstrate their generalization power compared to state-of-the-art techniques.

Internet of Things (IoT) services have been implemented for several network applications from smart cities to rural areas. However, there are many barriers to provide an efficient solution for the IoT service deployment underlying innovation SDN/NFV-based technologies. First, though an IoT service can flexibly deploy via virtual network functions (VNFs), a deployment scheme needs to solve the joint routing and resource allocation problem, which becomes more difficult than the traditional centralized cloud/datacenter solution due to distributed resources in the edge-cloud network. In addition, due to uncertain workloads in IoT services, static optimization solutions may not deal with uncompleted knowledge of the entire input, which is often given by assumptions, but unrealistic in current provisioning approaches. Aiming to address these issues, we model an online mechanism for the dynamic IoT service chain deployment to optimize the operational cost in a finite horizon. We propose a JOint Routing and Placement problem for IoT service chain (JORP) that can dynamically scale in/out the number of VNF instances. We then propose a learning method to efficiently solve JORP based on branch-and-bound (BnB). Our proposed learning mechanism can intelligently imitate the branching/pruning actions of BnB, and remove unlikely solutions in the search space based on the deep neural network model to improve the performance. In that respect, we take an intensive simulation that illustrates the promising result of our proposed deep learning method compared to BnB and the greedy baseline in terms of the performance of the algorithm and the operational cost reduction.

In the network function virtualization (NFV) environment, network services are usually provided in the form of service function chains (SFCs), which defines the link order of virtual network functions required in service requests and are mapped to the physical network. Although NFV facilitates the flexible provision of network services, service interruptions may occur as a result of software and hardware failures. Current solutions mostly use the backup method to ensure the reliability of SFCs. However, these methods ignore the SFC construction phase that has an impact on reliability. Besides, the resource efficiency still requires improvement. To address these issues, reliability-oriented SFC construction and backup problems are investigated in this work. First, an instance-sharing and reliable construction algorithm (ISRCA) is proposed to aggregate multiple SFCs into a service function graph (SFG), and perform reliability screening for the SFG set. After mapping the SFG to the physical network, a node-ranking algorithm with centrality and reliability (NRCR) is proposed for backup node selection and backup instance deployment to improve the reliability of SFCs that have not met the requirements. Experimental results demonstrate that under the premise of ensuring reliability, the proposed backup method can reduce the consumption of bandwidth resources by about 11.7%, when combined with the proposed construction method, it can further reduce the backup resources by 13.9%.

Hybrid electrical/optical (E/O) switching data center network (DCN) has recently emerged as a promising paradigm for future DCN architectures. However, there exist two major challenges: 1) the traffic is a mixture of both stable and burst components due to the diverse and heterogeneous user demands; 2) current scheduling algorithms are mostly static and not designed for the complex structure of hybrid E/O switching DCN, provoking frequent burst traffic congestion and performance degradation. This article endeavors to overcome the above challenges as follows. We first construct an error feedback-based spiking neural network (SNN) framework with high accuracy burst traffic prediction. We then design a prediction-assisted scheduling algorithm to handle the worst-case burst traffic. On the one hand, the error feedback-based SNN framework can significantly enhance the extraction of burst traffic features by mimicking the biological neuron system. On the other hand, prediction-assisted scheduling arranges the well-predicted traffic using a global evaluation factor and a traffic scaling factor. The simulation results reveal that our approach can efficiently integrate a spiking neural network into the traffic scheduling scheme and achieve satisfying performance with affordable computational complexity.

In the recent literature, Machine Learning (ML) techniques are increasingly used to detect the abnormal behaviour for different applications. Recently, these applications have moved to the cloud and virtualized environments due to the unique benefits such as deployment agility, scalability, flexibility and resiliency. However, those benefits pose a new challenge for classical ML approaches to accurately identify abnormal behaviours due to their highly dynamic and heterogeneous nature. In this paper, we propose a new approach Malchain for profiling virtual applications based on using a new concept: microservice role. The roles are used to provide a consistent view of the virtual application addressing the mentioned new challenges. The microservice data exchange graph built using this consistent view is then used to extract features providing the appropriate measures to profile the aggregated behaviour of the microservices comprising a virtual application. We show the efficiency and feasibility of our approach by implementing several different real-world attacks, and measuring high detection rates (86%-99%) for those attacks.

The following topics are dealt with: learning (artificial intelligence); optimisation; telecommunication computing; resource allocation; wireless channels; Internet of Things; telecommunication traffic; deep learning (artificial intelligence); mobile computing; cellular radio.

As an emerging solution to latency requirements of Internet of Things (IoT) services, edge computing can bring powerful processing capacity closer to data sources. However, with the limited resources at edge nodes, a major challenge is finding optimal resources in distributed edges to reduce the operational costs of service deployment. Prior works focus mainly on static optimization which may not work efficiently with the time-varying workloads and resource constraints. In this paper, we, therefore, consider a dynamic allocation framework in the edge-cloud network over the long run with uncertainty workloads. In such a system, we introduce a JOint Routing and Placement problem for IoT services, called JORP, that dynamically assigns resources according to workload demand in order to reduce the operational costs in long term. Inspired from the well-known algorithm, branch-and-bound (BnB), for solving the mixed-integer non linear problems (MINLPs) like JORP, we bring the learning concept to address the high complexity of BnB when the search space is huge. Particularly, we design a deep neural network (DNN) and train it under the imitation learning to mimic branching behaviors in BnB for searching the optimal solution. Finally, simulations show our solution outperforms baselines in terms of convergence and operational cost.

Optimized routing in multi-layer multi-domain (MLMD) IP-optical networks is challenging due to different technologies and policies in different domains. In this paper, we investigate the problem of using the hierarchical path computation engine (PCE) to leverage the performance of FlexE-the new flexible Ethernet technology which is used to map traffic between different layers and different domains. Our proposed PCE can be implemented in MLMD orchestration platforms to optimize network utilization while meeting delay constraints. We formulate the optimization problems of traffic routing and physical slot assignment for both FlexE-Aware and FlexE-Unaware modes with respect to QoS requirements, intra-domain information privacy and FlexE constraints. To solve the problem, we propose new algorithms that jointly optimize the routing and FlexE client assignment in polynomial time. To deal with the issue of missing intra-domain information, we use a novel implicit routing strategy to collect the intra-domain information from the child PCEs. Simulation results show the proposed solution achieves 90.1% higher efficiency than the state-of-the-art solutions.

A main component of the Multi-Layer Multi-domain (MLMD) orchestration is the end-to-end path computation over the packet and optical layers. Routing in MLMD networks is complex and requires special computational elements and cooperation between different layers and domains. The goal is to optimize the utilization of Wide Area Networks (WANs) leveraging on FlexE - the new Flexible Ethernet technology which couldn’t be fully achieved from local resource allocation in a single domain. We present MLMD-PCE, a path computation engine for MLMD networks that achieves optimal routes through a hierarchical path computation. We formulate an optimization problem of traffic routing and resource assignment for FlexE-Aware and FlexE-Unaware modes and propose an approximation algorithm that runs in polynomial time. Another issue of multi-domain routing is the lack of visibility over the intra-domain typologies in the parent-PCE. To solve this problem, we use a novel mechanism to gather the intra-domain information from the child-PCEs while keeping the domain privacy. Simulation results show that MLMD-PCE carries 77% more traffic than the current Hierarchical-PCE.

This paper investigates the problem of resource allocation for multiple providers on a heterogeneous edge-cloud environment. Unlike prior work that tried to solve this problem using traditional optimization, we model the problem as a non-cooperative game and show it is closely similar to a transportation game. We chose this method to deal with our multi-objective function problem which is not suitable with traditional one as the latter generates a single optimal solution. We propose a solution to the game based on Removal of Strictly Dominated Strategy (RSDS) and Best Response Methods (BRM). The goal is to seek the Nash equilibrium over the generated set of equilibria. Our simulation result shows that our proposed solution can find a set of minimized equilibria and an equilibrium point in most of the executions, which outperforms other methods using only RSDS.

The unmanned aerial vehicle (UAV) cellular communications is a promising technique to boost coverage and realize intelligent applications. The communication power consumption is a major challenge faced by the UAV cellular networks. In this work, we consider a single UAV user equipment (UE) connected with a terrestrial base station (BS), and focus on the uplink (UL) performance. Through theoretical analysis, we find an optimal power control factor in terms of the communication energy efficiency, when the distance between the UAV and the BS is small.

This paper presents an energy-saving wireless power allocation scheme based on convex optimization (PA-CO) in cache-enabled Fiber-Wireless (FiWi) access networks. Experiments indicate that the proposed PA-CO improves the global throughput when confronted with large-scale access terminal sets in 5G IoT scenario, while remaining total power consumption to a low level.

In this paper, a cumulative distribution function (CDF)-based scheduling scheme for uplink non-orthogonal multiple access (NOMA) network is investigated. With considering imperfect successive interference cancellation (SIC) and SIC power constraint, closed-form expressions for the outage probability of two scheduled users are derived in cognitive-radio-inspired power allocation (CPA) scenario. To get more insights, high SNR approximations of the outage probabilities are given, and the results reveal that the two users can achieve a diversity order linear with the number of users. Simulation results validate the accuracy of the analytical expressions.

Due to the limited energy in massive machine type communication (mMTC), improving energy efficiency is necessary for cellular IoT transmission. This paper considered the edge IoT devices to access relay nodes and used NOMA scheme to transmit information to improve system energy efficiency and reliability. Firstly, we proposed a multi-hop uplink NOMA assisted IoT transmission model with rely cooperation. The transmission model uses NOMA technology and needs to control the power. Secondly, for the channel matching problem, traditional algorithm models such as KM and Hungarian algorithms are only applicable to two-hop system. The paper proposed a multi-hop channel allocation model to improve transmission reliability. Then we used an intelligent optimization algorithm to solve optimization problem. This paper improved the GSO algorithm and proposed the GCM algorithm. GCM can solve the problem of local optimization of traditional GSO algorithm. Finally, in the simulation part, we compared the GCM algorithm with GSO, ACO and random matching. The GCM algorithm has higher reliability, which is 8% and 12% higher than GSO and ACO. It is clear that energy efficiency achieved by GCM is 10% and 18% than GSO and ACO, respectively.

Internet of Things (IoT) has attracted tremendous interest since it can improve production efficiency and system intelligence significantly. However, with the explosive growth of various types of device and data flow, IoT suffers spectrum resource scarcity for wireless applications. In this paper, we propose a solution for spectrum resource sharing in the IoT network, with the objective to facilitate the limited spectrum sharing between different kinds of sensors. To overcome the challenges of unknown dynamic IoT environment, the deep Q-learning network (DQN) is adopted. BS acts as the single agent and centrally manages all spectrum resources. First, a new reward function is designed to drive the learning process, which takes into account the different communication requirements of various sensors. In addition, to improve the learning efficiency of DQN, we compress the action space. Finally, simulation results show that compared with other algorithms, the proposed algorithm can achieve good network performance.

This paper demonstrates an ultra-high precision time synchronization (U-TS) scheme by reducing link asymmetry for cloud radio over fiber network (C-RoFN) in beyond 5G, The U-TS scheme is supported by a deep reinforcement learning (DRL) based autonomous synchronous signal routing algorithm. Experimental results show that the proposed U-TS scheme achieves <; 100 ns synchronization accuracy by using a large realistic dataset.

Network slicing is a widely discussed technology for satisfying the diverse requirements in 5G and beyond scenarios. It enables the operators to create end-to-end virtual slices in network infrastructures. The flexible assignation of multi-dimensional slice resources including radio and spectral resources is essential for 5G communication networks. However, there is a lack of effective solutions to reconfigure the spectral and the radio resources in 5G radio access network (RAN) and core network. In this paper, we realize orchestration functionalities by exploiting a novel two-step slice reconfiguration strategy. Firstly, a slice-monitoring model is proposed to map slices to vectors by representation learning. Secondly, a slice reconfiguration algorithm is introduced to realize flexible slice reconfiguration. The slice reconfiguration strategy can decide when to trigger the reconfiguration strategy and how to reconfigure the slice resources. As for network performance, simulation results show our reconfiguration strategy can further improve slice resource utilization rate by 31% as well as reduce the blocking rate by 40% in 5G RAN and core network.

Smart home gateway has to process different types of network traffic generated from several devices in an optimal way to meet their Quality-of-Service (QoS) requirements. However, the fluctuation of network traffic distributions results in packets concurrency. Current QoS-aware scheduling methods in smart home networks do not consider concurrent traffic in their scheduling solutions. This article presents an analytic model for a QoS-aware scheduling optimization of concurrent smart home network traffic with mixed arrival distributions and using probabilistic queuing disciplines. We formulate a hybrid QoS-aware scheduling problem for concurrent traffic in smart home network, propose an innovative queuing design based on the auction economic model of the game theory to provide a fair multiple access over different communication channels/ports, and design an applicable model to implement auction game on both sides; traffic sources and the home gateway, without changing the structure of the IEEE 802.11 standard. Our experiments show the proposed solution achieves an improvement of 14% of packets that meet their required delay and 57% of delay for different number of concurrent flows in the system.

In this paper, we propose a novel method for Multispectral document image binarization (MSdB) through the Non-negative Matrix Factorization (NMF) approach. We propose a three-step MSdB-NMF framework: i) NMF-based feature extraction algorithm by introducing a new optimization problem; ii) post-processing method iii); apply any existing gray/RGB binarization scheme. In the first step, we extract N features out of B spectral bands (N < B) and their corresponding coefficient matrix. We introduce a novel objective formulation that considers the robustness (related to the noise and various types of degradations) and sparseness (related to the ratio of text pixels versus the background). We employ the multiplicative updating rules to solve the proposed minimization problem and prove the convergence of the proposed feature extraction algorithm. In the next step, we select an appropriate feature vector, equivalently the corresponding coefficient vector. We propose to select it either visually or automatically via a post-processing method, which uses the benchmark binarization methods as baseline. In the last step, we apply some existing binarization methods such as Sauvola and Howe over the selected coefficient vector. Our proposed binarization framework is applicable for any kind of MS or hyperspectral (HS) document image without considering any prior knowledge such as the side information about the spectral bands of MS/HS document image. We evaluate our proposed binarization framework over two MS document image datasets. The experimental results confirm that our proposed framework outperforms several state-of-theart binarization schemes including the winner of the contest in MS-TEx-2015.

This paper investigates an efficient placement and chaining of Virtual Network Functions (VNFs) to provide cloud based IoT services with minimal resource usage cost. We take into account bandwidth capacity and link delay of network connection between clouds where VNFs are allocated and underlying IoT networks where sensors and IoT gateways are deployed. Regarding the constantly changing network dynamics, input traffic of service components is considered at the lower granularity level of messages based on the communication between each VNF and corresponding sensors via IoT gateways. From the algorithm perspective, the specific topology of multiple edge clouds is leveraged to improve the solution. In this paper, we present an NFV-based high-level architecture for a system that enables the deployment of IoT services across multiple edges and clouds. We formulate the VNF placement problem using a non-convex Integer Programming model. Taking into account different IoT topologies, we devise two algorithms for small- and large-scale networks to find the near optimal solution: i) a customized Markov approximation with two techniques, i.e., multi-start and batching, and a node ranking-based heuristic. Simulation and experimental results show that the proposed solution improves the cost up to 21% compared to state-of-the-art schemes.

Smart homes play a crucial role in reducing the residential sector electricity consumption and Greenhouse Gases (GHG) emissions. In this work, we present a time series approach to predict GHG emissions to be integrated into smart home management systems. More specifically, we used Long Short-Term Memory (LSTM), a variant of Recurrent Neural Networks. The prediction results get mean absolute percentage error (MAPE) close to 2 percent when the region under study has an energy matrix mostly based on fossil fuels, less intermittent. For regions in which more renewable sources are present, the MAPE is around 12 percent. However, in either case, LSTM can predict the hours well with smaller emissions among the next 24 hours. Such day-ahead information brings awareness to the users and allows the scheduling of appliances to work in the hours in which the emissions are minimal, reducing them without significantly affecting the consumers' behavior.

Smart home gateway has to process different types of network traffic generated from several devices in an optimal way to meet their QoS requirements. However, the fluctuation of network traffic distributions results in packets concurrency. Current QoS-aware scheduling methods in the smart home networks do not consider concurrent traffic in their scheduling solutions. This paper presents an analytic model for a QoS-aware scheduling optimization of concurrent smart home network traffic with mixed arrival distributions and using probabilistic queuing disciplines. We formulate a hybrid QoS-aware scheduling problem for concurrent traffics in smart home network, and propose an innovative queuing design based on the auction economic model of game theory to provide a fair multiple access over different communication channels/ports. Our experiments show the proposed solution achieves an improvement of 14% of packets that meet their required delay and 57% of delay for different number of concurrent flows in the system.

Software in IoT devices needs to be improved regularly to adapt security issues and new user requirements. In advanced IoT networks, devices employ the component-based software architecture in which components can be updated at run-time, such devices can download software components from neighbors, enabling fast distribution of updates in the entire network. One of the most energy consuming operations in the update process is flash re-writing in which the order of re-writing components into the flash memory is decisive for energy consumption. In this paper, we propose a mechanism that schedules updates in an entire IoT network to minimize the energy consumption, while satisfying the deadline constraint for updating all the devices. We mathematically formulate the problem of energy efficient update scheduling as an optimization problem with a novel energy model of the update process, then propose an algorithm to approximate the optimal schedule for updating all devices in the network. We examine the proposed algorithm in three different network instances including a tree, a partial mesh and a full mesh topology. Simulation results illustrate that our algorithm can obtain a near optimum which is, in the best case, only 3.2% different from the minimum.

The emerging paradigm of Software Defined Network (SDN) and virtualization technology promises an efficient solution for network providers to deploy services. Adopting them not only facilitates network management but also helps reduce the cost of maintaining network infrastructure. However, despite these advantages, there are still obstacles that must be overcome before SDN and virtualization can advance to reality in industrial deployments. In this paper, we focus on two well-researched issues, namely controller-switch assignment and Virtual Network Function (VNF) placement. Unlike prior works, our purpose is to jointly solve these two problems, accounting for the complex and counter-intuitive manner they are related to each other. We present a service chain aware framework (SACO) that enables the controller-switch association in a multi-controller network regarding the relationship of switches via their connected VNFs that implement service components of the chain. We also propose a model and formulate the joint optimization problem of dynamic controller-switch mapping and VNF allocation. We apply the Lyapunov optimization framework to transform a long-term optimization problem into a series of real-time problem and employ the Markov approximation method to find a near-optimal solution. Simulation results show that our service chain aware approach improves the system cost up to 10 ~ 43% compared to the state-of-the-art solutions.

Due to security issues and incremental user requirements, software in IoT devices needs to be changed frequently. Recently, advanced IoT devices employ the component-based software architecture in which components can be updated at run-time. In such IoT networks, devices can download updated components from neighbor nodes, enabling quick deployment of updates in the entire network. A key operation which consumes a significant amount of energy in the update process is flash re-writing, in which the order of re-writing components into the memory is decisive for energy consumption. In this paper, we propose a mechanism that schedules updates on all devices in an IoT network to minimize the energy consumption, taking into account the deadline constraint for updating the entire network. We introduce a novel energy model of the update process, then propose an algorithm to approximate the optimal schedule for updating all devices in the network. Simulation results show that our algorithm can obtain a near optimal which is, on average, 7.1% different from the global minimum.

In this paper, we propose a novel Blind Source Separation (BSS) based framework for multispectral (MS) document images binarization. This framework takes advantage of the multidimensional data representation of MS images and makes use of the Graph regularized Non-negative Matrix Factorization (GNMF) to decompose MS document images into their different constituting components, i.e., foreground (text, ink), background (paper, parchment), degradation information, etc. The proposed framework is validated on two different real-world data sets of manuscript images showing a high capability of dealing with: variable numbers of bands regardless of the acquisition protocol, different types of degradations, and illumination non-uniformity while outperforming the results reported in the state-of-the-art. Although the focus was put on the binary separation (i.e., foreground/background), the proposed framework is also used for the decomposition of document images into different components, i.e., background, text, and degradation, which allows full sources separation, whereby further analysis and characterization of each component can be possible. A comparative study is performed using Independent Component Analysis (ICA) and Principal Component Analysis (PCA) methods. Our framework is also validated on another third dataset of MS images of natural objects to demonstrate its generalizability beyond document samples.

The last few years have witnessed a significant increase in the deployment of heterogeneous Internet of Things (IoT) networks. IoT devices send data with different requirements such as tolerated delay and data rates. Emerging multi-interface IoT devices bring the flexibility of connecting to multiple heterogeneous access networks, which thus improves the network capacity. However, each network interface has its own constraints in terms of network coverage, capacity, packet loss rates, etc. An efficient utilization of the available multiple interfaces in IoT gateways would improve the network performance. Therefore, it is crucial to design a flow assignment mechanism to select the appropriate interface that best satisfies the flow's requirements and maximizes the amount of data accepted by an IoT gateway. In this work, we model and formulate the optimized flow assignment problem (OFAP) in a multi-interface IoT gateway. Then, we develop two heuristic algorithms to find a feasible solution for OFAP. The first algorithm is based on the greedy approach and the second uses dynamic programming to assign flows to interfaces. We provide simulation results that show the effectiveness of our algorithms.

Smart home gateways have to forward multi-sourced network traffic generated with different distributions and with different Quality of Service (QoS) requirements. Most of the current QoS-aware scheduling methods consider only the conventional priority metrics based on the IP Type of Service (ToS) field to make decision for bandwidth allocation. Such priority-based scheduling methods are not optimal to provide both QoS and QoE (quality of experience) since higher-priority traffic do not necessary require higher stringent delay than lower-priority traffic. To solve the gaps between QoS and QoE, we propose a new queuing model for QoS-level Pair traffic with mixed arrival distributions in Smart Home network (QP-SH) to make a dynamic QoS-aware scheduling decision which meets delay requirements of all traffic while preserves their degrees of criticality. A new metric which combines, the ToS field and the maximum number of packets that can be processed by the system's service during the maximum required delay, is defined. Our experiments show the proposed solution provides an improvement regarding the number of packets that meet their priorities and their maximum delays as well as the mean number of packets in the system.

Adaptive Link Rate (ALR) is widely used to save energy consumption of network by adjusting the link rate according to the carried traffic through a network-level optimization of the flow allocation process. Existing ALR solution is mainly reactive, in which link speed is changed only when new traffic demand is requested. Also, they focus on energy consumption, and do not consider the cost of changes in the network (e.g., change in traffic routes, and link rates). Once bandwidth has been allocated for a demand, the link rate remains constant during the entire session. Therefore, this solution may result in sub-optimal schemes and requires multiple re-optimizations as traffic flows are fluctuating during the session, hence reducing the overall network performance. In this paper, we improve the ALR with a multiple-step-ahead method to optimize link rates based on forecasting traffic demand predictively. We formulate the proposed Predictive ALR (PALR) as an Integer Linear Programming (ILP) model and then design a heuristic simulated annealing (SA) -based algorithm to solve it. Our experimental results show our approach provides energy saving while it decreases on average 18% of link state transition and 11% of the flow reroutings compared to the original ALR.

In the future Fifth-Generation networks, the eavesdropping is a critical threat due to their broadcast-based transmission. This problem can be addressed with the cryptographic protocols. However, this method is complex and difficult because of the dynamic topology of wireless networks, which does not allow an efficient management of security keys. As a complement solution, Physical-layer security (PLS) is integrated to enhance secrecy in wireless networks. The PLS exploits the schemes features of this layer, namely the modulation, Massive Multi-Input Multi-Output(m-MiMo) and channel coding. The fountain code is one of those systems where the secrecy is provided when the destination retrieves packets encoded before the intruder. Nevertheless, the secrecy can not be guaranteed when eavesdropper uses large number of the antennas as in the m-MiMo. The feature of m-MiMo should be considered to secure main channel with fountain codes. Therefore, we propose to use Raptor code which is a class of fountain code, aided by an Artificial noise (AN) and the punctuated data to reduce the efficient of intruder channel. This allows the main channel to retrieve the signal before eavesdropper. The numerical results show that using Raptor code in massive MiMo enhances the reliability and the security on the channel of legitimate user, while minimizes the abilities of intruders to spy on data.

Raising density and ever-increasing traffic demand within future 5G Heterogeneous Networks (HetNets) will result in huge deployment, expansion and operating costs for upcoming Mobile Backhaul (MBH) networks. Multi-tenancy and network slicing based on virtualized resources are promising solutions to satisfy MBH network greediness while reducing related expenditures. Nevertheless, there is no appropriate model that fairly distributes costs over multiple Mobile Network Operators (MNO), and also optimizes physical resource planning. In this paper, we introduce a new model of 5G multi-tenant MBH costs (CapEx and OpEx). Then, we drive a novel pay-as-you-grow and optimization model called Virtual-Backhaul-as-a-Service (VBaaS) as a planning tool optimizing the Project Profit Margin (PPM) while considering the Total-Cost-of-Ownership (TCO) and the yearly generated Return-on-Investment (ROI). We also formulate an MNO pricing game (MPG) for TCO optimization to calculate the optimal Pareto-Equilibrium pricing strategy for offered Tenant Service Instances (TSI). Finally, we compare the PPM for a specific use-case known in the industry as CORD project using Traditional MBH (T-MBH) versus Virtualized MBH (V-MBH) as well as using randomized versus Pareto-Equilibrium pricing strategies. Numerical results show more than three times increase in network profitability using our proposed solutions compared with Traditional MBH (T-MBH).

This paper investigates an efficient placing mechanism for placing Virtual Network Function (VNF) on the cloud networks to enable the construction of Internet of Things (IoT) service chains with minimal resource usage cost. In particular, we propose a model taking IoT network infrastructure into account. Such the network composed of numerous sensors with constrained resource, dynamic connectivity toward multi-homing IoT gateways makes the problem of optimally placing VNFs with expected Quality of Service (QoS) more challenging and has not been considered in prior works. From the model, we formulate a non-convex Integer Programming (IP) placement problem and devise a batching Markov approximation placement (BMAP) algorithm to find the optimal solution. Simulation results show that the proposed approach improves the cost compared to those that do not consider the IoT network.

With the rapid evolution of smart city all over the world, the appealing services of IoT and big data analytics have prompted the design of more reliable assurance mechanism for network quality. It has been a crucial issue of network operation that once multiple links fail simultaneously, the transmission of real-time services cannot be guaranteed. Therefore, rapid locating of faults is the premise for network to recover quickly. However, current faults location methods can't satisfy the requirement due to the expansion scale of wireless and optical networks and the growing demands of customers. In this paper, we propose an efficient multi-link faults location algorithm based on Hopfield Neural Network (HNN). We make full use of the information of network topology and the services transmitted to model the relationship between fault set and alarm set. HNN is used as an optimization method to analyze the uncertainty of faults and alarms and to find where the faults most likely occur by constructing a proper energy function. It has been proved by experiments that this method can achieve real-time faults location while ensuring positioning accuracy, which provides a good solution for smart city service assurance.

Network functions virtualization intertwined with software-defined networking opens up great opportunities for flexible provisioning and composition of network functions, known as network service chaining. In the cloud, this allows providers to create service chains tuned to each application type while optimizing resources' utilization. This is particularly useful to accommodate different tenants' applications with different security needs. However, considering security provisioning from the single perspective of resources optimization may lead to deployment solutions that do not comply with well-known security-related best practices and recommendations. In this paper, we propose network security defense patterns (NSDP) aimed at leveraging the best practice and know-how from the security experts and at capturing various security constraints to efficiently select compliant security functions' deployment options. The placement problem being a NP-Hard problem to solve, we also propose a scalable networking and computing resources aware optimization framework to efficiently provision different NSDPs. We further show the feasibility of implementing NSDPs in the cloud infrastructure through the integration of our approach into an open source cloud framework, namely OpenStack, in our test laboratory. The simulation results show the effectiveness of our approach in selecting an optimal placement of the security functions for large data centers with hundreds of thousands of computing nodes, while complying with the predefined security constraints and improving the scalability compared to the current placement algorithms.

Accelerated Software Defined Networking (SDN) adoption makes SDN paradigm emerging in the state of the art. Especially, the combination of SDN and network functions virtualization (NFV) becomes a promising trend in deploying virtual network services for network operators. Although SDN can decouple networks into the control plane and the data plane to obtain flexible operation and programmability, there are many open issues that need to be addressed for SDN deployments, such as i) where to place SDN controllers in a given network, ii) how to assign connections from controllers to switches in terms of satisfying multiple objectives (e.g., resource utilization, failure, quality of services, etc.). In this work, we focus on the efficient assignment between SDN controllers and switches to guarantee a low operational cost, quality of services, and fault tolerance, in which the complexity of virtual links in network services, an omitted factor in current works, is considered and addressed. We formulate an optimization problem for dynamic controller/switch mapping (DCSM) in the network virtualization. We then proposed approximation algorithms in terms of relaxing the binary variables to solve the NP-hard problem, DCSM, in both centralized and distributed mechanisms. We also create various simulation schemes to evaluate our methods where they outperform state-of-the-art methods.

In this paper, we propose a hybrid approach of two efficient 2-D stochastic models: the classical NSHP-HMM (Non-Symmetric Half-Plane Hidden Markov Model) and the NSHPZ-HMM based on conditional zone observation probabilities. The main advantage of the hybrid model is its ability to properly model an image through a multi-level analysis of features (low-level and high-level information) and interpretation (pixel-based and zone-based views). First experiments conducted on scaled digit images of MNIST show that the proposed hybrid model can improve the recognition accuracy over the two baseline models.

Separating content from noise in historical manuscripts is a fundamental task in digital palaeography. This study presents a fully automated segmentation approach based on the response of Harris corner detectors. The strength and clustering efficiency of the detected corners in the manuscripts are evaluated and used to segment the content from the background and noise. In addition, a manuscript reconstruction technique is proposed from the gradient field using the Poisson method to guide the interpolation. This reconstruction is able to remove noise significantly and hence enhances the contrast of the content thus making it easier for users to read and process these documents. The proposed approaches are evaluated using various standard databases to highlight their effectiveness and robustness to a multitude of noise and writing styles. Subjective and objective evaluations of the experimental results show that these techniques are able to successfully segment and reconstruct a very diverse set of scanned documents. An analysis of the results has also shown that the proposed technique compares favourably against similar counterparts.

Designing an efficient defense framework is challenging with respect to a network's complexity, widespread sophisticated attacks, attackers' ability, and the diversity of security appliances. The Intrusion Response System (IRS) is intended to respond automatically to incidents by attuning the attack damage and countermeasure costs. The existing approaches inherit some limitations, such as using static countermeasure effectiveness, static countermeasure deployment cost, or neglecting the countermeasures' negative impact on service quality (QoS). These limitations may lead the IRS to select inappropriate countermeasures and deployment locations, which in turn may reduce network performance and disconnect legitimate users. In this paper, we propose a dynamic defense framework that selects an optimal countermeasure against different attack damage costs. To measure the attack damage cost, we propose a novel defense-centric model based on a service dependency graph. To select the optimal countermeasure dynamically, we formulate the problem at hand using a multi-objective optimization concept that maximizes the security benefit, minimizes the negative impact on users and services, and minimizes the security deployment cost with respect to the attack damage cost.

The emerging paradigm of network function virtualization (NFV) technology promises an efficient solution for optimized service deployment in the cloud computing environment thanks to its ability to dynamically add or remove virtual resources when there is a change in workload. Nevertheless, telecom providers are still facing a challenging issue in efficiently adopting NFV to deploy Web real-time communication (WebRTC) service on top of IP multimedia subsystem (IMS). Providing WebRTC service increases the inherent complexity of the IMS system in terms of the number of service nodes as virtual network functions (VNFs) and the way they interact, both of which play significant roles in the problem of optimally allocating resources. This paper proposes a virtualized interworking system between IMS and WebRTC called NFV-based interworking architecture, and describes the mechanism for VNFs to exchange messages with each other. We present an analytic system model considering the constraints of resources, quality of service (QoS), and service costs. A real-time Markov approximation-based resource allocation algorithm (RIDRA) is then designed allowing a provisioned resource at service nodes to be reconfigured in time to meet performance requirements. The proposed solution is evaluated on the large scale by simulation and on the small scale by our developed testbed. Experimental results reveal that our algorithm effectively responds to fluctuating service demands with a service cost reduced by 19% via efficiently allocating virtual resources while maintaining QoS requirement.

With the introduction of various hardware/software technologies such as Cloud Technologies or Virtualization technologies, there has been a great potential to reuse ICT resources. These technologies also provide various advantages including resource consumption reduction (in the use phase only or in the whole life cycle). In addition, there are additional potentials regarding the resource opacity introduced by these technologies. These potentials could improve cloud solutions and enable tweakability of components' nonessential features that may be not captured in the component models. This is especially of great advantage in the case of heterogeneous/disaggregated infrastructure where developing comprehensive models to cover everything is practically impossible. In this work, by leveraging on the concept of pathways, we develop a few mechanisms that enable transparency and therefore tweakability of features even in the presence of abstraction and heterogeneity. In particular, the layered-stack approach to system decomposition is considered because of its association with software defined networking (SDN). For a concrete example, the case of dynamic frequency scaling of processors is considered and it is shown that the associated consumption could be considerably reduced without requiring additional changes to the middle-layer components.

Many applications are distributed in cloud as they consist of different modules and tiers. Cloud tenant must be able to monitor the deployed applications to ensure that it is operating correctly, meeting its SLAs, and fulfilling business requirements. Activating all type of monitoring functions for all tenant's applications at the same time is costly. The more required monitoring functions the more allocated cloud resources. Moreover, it incurs monetary cost for tenants. In this paper, the problem of virtual monitoring function (vMF) placement for service chains is modeled by maximizing both the network and computing resource utilization. Beside that, we propose a set of tenant policies, which are either monetary-driven or performance-driven, translated as constrains in optimal placement algorithm. Simulation results show that the proposed model can reduce the total incurred cost by up to 10% compared to the best non-optimal heuristic. Moreover, the proposed model can decrease the execution time about 84% compared to the basic solution.

Word spotting in document images is a challenging problem, due to the large intra-class variability in handwritten shapes and the lack of labeled data. To tackle these challenges, this paper proposes an efficient multiscale representation for word images, which is learned in an unsupervised manner using the spherical k-means algorithm. A pooling function is applied in a spatial grid to obtain a fixed-length vector of features, robust to small shifts in the image. Scale variability in handwritten data is also considered by using patches of various sizes in the encoding process. Another important contribution of this work is to model the training-based word spotting task as a classification problem in the space of representation differences, thereby allowing the learned model to find matches for word classes that were not seen in training. The proposed system is evaluated on the well-known George Washington (GW) dataset. Experimental results show that our system outperforms state-of-the-art word spotting approaches in both training-free and training-based scenarios.

In this paper, we will study the capacity of 5G cellular networks with caching. We first investigate the capacity using uncoded caching and then we study the capacity of such networks using a novel technique based on file categorization and coded cache placement. In our cache placement strategy, we propose to create file groups with approximately similar popularity levels (within a factor of two) and then we randomly code the files in each file group together and place them in the User Terminal (UT) caches. We prove that this technique can significantly increase the network capacity by reducing the average traffic. Our proposed coded cache placement strategy is decentralized and performs close to optimal. Simulation results also confirm our theoretical results.

Although Internet of Things (IoT) has been growing rapidly in recent years, the deployment of large-scale IoT solutions are facing several challenges, in particular resource optimization. This is due to high requirements of IoT applications which are usually composed of numerous services each of which could be provided by multiple providers with various specifications. Since an IoT service is often micro-services, dynamically opting appropriate micro-services to guarantee the quality of delivered services to end-users is challenging because conventional network paradigm was not designed to support removal or insertion of modular services on the fly. In this paper, we investigate a cloud-based IoT service orchestration architecture to facilitate the IoT service deployment and propose algorithms to optimize service chaining process. Experimental results show our algorithms effectively orchestrating microservices by reducing total service cost up to 36% while maintaining real-time requirement.

Smart home applications are becoming increasingly popular due to the explosive growth of intelligent devices in recent years. However, such applications require extra computing resources. In this paper, the task offloading management in the smart home scenarios are investigated in micro and macro scales. First, the optimal process of different applications is determined to be offloaded from the smart home controller to the cloud system. Second, a policy in the home gateway is proposed to select the best server in the cloud to dynamically offloading the computing tasks considering energy saving while satisfying the applications QoS metrics. Using queuing theoretical approach, we model the smart home application execution time at macro level considering the applications QoS constraints. Then, we apply duality theory to find the optimal offloading ratio. Moreover, an optimal interface/server selection is investigated to minimize the energy consumption. On the other hand, Lyapunov drift optimization is applied at micro level to find the optimal performance point of the smart home applications in real time. The effectiveness and optimality of our proposed algorithms are evaluated by experiments over realistic smart home scenarios.

Information and communication technologies are believed to contribute to reduce environmental impacts in other economic sectors. Smart homes aim to this purpose through improved resource management, but actual environmental benefits are uncertain. Indeed, an increase in efficiency may lead to an increase in consumption and thus in environmental impacts, a phenomenon called rebound effect. Here we describe an agent-based approach which simulates human behaviors during the use phase of smart homes and allows computation of environmental impacts including those due to rebound effect. In our simulation, smart homes reduce environmental impact by 2%. However, rebound dims between 6% to 24% of the smart homes' environmental benefits depending on impact category. Moreover, a sensitivity analysis indicates that weather and electricity mix variability should be considered when designing policies related to smart homes technology. Life cycle assessment by this method accounts for user behaviors, which is necessary when studying some complex systems found in sustainable consumption. Evaluate the environmental performance of a smart-city and study different rebound mechanisms are possible applications of the developed approach.

Unconstrained offline handwriting recognition is a challenging task in the areas of document analysis and pattern recognition. In recent years, to sufficiently exploit the supervisory information hidden in document images, much effort has been made to integrate multi-layer perceptrons (MLPs) in either a hybrid or a tandem fashion into hidden Markov models (HMMs). However, due to the weak learnability of MLPs, the learnt features are not necessarily optimal for subsequent recognition tasks. In this paper, we propose a deep architecture-based tandem approach for unconstrained offline handwriting recognition. In the proposed model, deep belief networks are adopted to learn the compact representations of sequential data, while HMMs are applied for (sub-)word recognition. We evaluate the proposed model on two publicly available datasets, i.e., RIMES and IFN/ENIT, which are based on Latin and Arabic languages respectively, and one dataset collected by ourselves called Devanagari (an Indian script). Extensive experiments show the advantage of the proposed model, especially over the MLP-HMMs tandem approaches.

In this letter, a novel decolorization method is proposed to convert color images into grayscale. The proposed method, called CorrC2G, estimates the three global linear weighting parameters of the color to gray conversion by correlation. These parameters are estimated directly from the correlations between each channel of the RGB image and a contrast image. The proposed method works directly on the RGB channels; it does not use any edge information nor any optimization or training. The objective and subjective experimental results on three available benchmark datasets of color to gray conversion, e.g., Cadik, CSDD, and Color250, show that the proposed decolorization method is highly efficient and comparable to recent state-of-the-art decolorization methods. The MATLAB source code of the proposed method is available at: http://www.synchromedia.ca/system/files/CorrC2G.m.

In mobile web conferencing, slide decks should be optimized before delivery to meet the constraints and environments of target mobile devices. To deliver optimally adapted slides, a trade-off between the visual aspect and delivery time must be reached. Static adaptation methods are CPU-intensive, and require large storage space. The dynamic approach is attractive as the optimal version is created on the fly when the actual slide is to be shared. Existing dynamic solutions are optimized for the resolution of the target mobile device and use good visual quality settings. However, they do not control the resulting data size, which creates serious usability issues, such as increasing the delivery time. Prediction-based methods require much less memory and processing resources than static approaches while yielding an excellent user experience. In this paper, we propose a multi-objective dynamic content adaptation framework, in which we maximize the visual quality and minimize the delivery time simultaneously. We compare our solution with an ideal optimal point, called utopia, and with all the optimal solutions (Pareto front) provided by a static exhaustive system. The obtained results show that our framework yields solutions very close to the utopia and, for the majority of the documents tested, the obtained solutions are on the Pareto front.

Among the innovative approaches to reduce the greenhouse gas (GHG) emissions of data centres during their use phase, electrical power from renewable sources appears promising. However, renewable electricity is often intermittent due to meteorological conditions. Consequently, the regional availability of renewable power varies constantly over time. This created the opportunity to deploy cloud computing systems relying on data centres located in different regions. Cloud computing technology enables real-time load migration to a data centre in the region where the GHG emissions per kWh are the lowest. While this approach is becoming popular to manage distributed data centres, there is still room for improvement in its implementation. Indeed, the consequences of data centre power demand migrations across electric networks and the resulting GHG emissions are usually neglected. In this project, we developed a novel GHG emission factor based on the sources of electricity affected by the server load migrations. Then, we used this emission factor in a simulation of distributed data centres to minimize their GHG emissions. Results show, the use of the novel emission factor enables an extra reduction of 23% of GHG emissions as compared to the usual approach.

Diversity, dynamicity, and the huge volume of traffic in the network between datacenters has risen network administrators concerns on how to efficiently visualize their system in real-time. To deal with these challenges, we present in this paper a visualization framework based on advanced machine learning, traffic characterization, sampling, and graphical visualization algorithms, which aims to efficiently support inter-datacenter network monitoring. Experimental results show the framework is able to process real-time big flows and provides human-friendly interactive graphical representations.

Despite recent efforts and important results already achieved, the reduction of energy consumption and carbon emissions by Information and Communication Technologies is still far from the expected goals. As the annual growth in traffic is doubling every two years with more and more connections to the Internet, to be energy and carbon-aware it is paramount to implement a Monitoring and Measurement System which supports green strategies in a geographically distributed environment. Such an environment has some specific challenges that must be taken into account, such as the WAN connection, security and latency concerns. On the other hand, it also provides opportunities to reduce operational costs and emissions, improve reliability and resources management etc. This work proposes a framework which is capable of supporting green metrics in network monitoring. The framework comprises temporally differentiated data on emission factors and provides ground information able to support different applications. We have implemented the framework in a nationwide testbed and our experiments show the framework is able to provide the ground information for customizable green metrics, like power/energy, traffic, and carbon equivalent emissions. This framework can be used as a support for a variety of applications which depend on energy and emissions metrics.

Cloud computing offers provision for elastic and scalable infrastructure resource allocation across the network that allows deployment of services for controlling home devices and appliances. Data generated from heterogeneous smart home devices are processed in different application services deployed in the cloud data center. The primary challenge of smart home service provider's is to optimize the cloud resource allocation while satisfying the Quality of Service(QoS) constraints of the application services. Service execution time is one of the most vital QoS parameters. In this paper, a queuing theoretic approach is proposed to model the smart home workload. First, M/M/c queue model is applied to find the response time of smart home tasks with light variation over the arrival rate. Then, Markovian Modulated Poisson Process (MMPP) is used to extend the model to a more advanced type of smart home processing workloads. Next, the optimal number of Virtual Machines(VMs) required deploying the application servers that can satisfy the execution time constraint of incoming workloads is calculated. Finally, total service time of a smart home application is calculated considering into account the possible level of concurrency and dependency among tasks of an application service. In the end, some numerical and simulation examples are provided to validate our findings.

The exponential growth of network users and their communication demands have led to a tangible increment of energy consumption in network infrastructures. A new networking paradigm called Software-Defined Networking (SDN) recently emerged which simplifies network management by offering programmability of network devices. SDN assists to lower link data rates via rate-adaptation technique which reduces power consumption of the network. The main idea behind this paper is to find a distribution of traffic flows over pre-calculated paths which allow adapting the transmission rate of maximum links into lower states. We first formulate the problem as a Mixed Integer Linear Program (MILP) problem. We then present four different computationally efficient algorithms namely greedy first fit, greedy best fit, greedy worst fit and a meta-heuristic genetic algorithm to solve the problem for a realistic network topology. Simulation results show that the genetic algorithm consistently outperforms the three greedy algorithms.

Evolving nature of network traffic challenges existing models to fit and predict its behavior. In particular, real traffic modeling requires more flexible design that can adapt to long-range and short-range dependent traffic with dynamic patterns. Unfortunately, existing models cannot handle such requirements because various traffic behaviors such as periodic and self-similar are not taken into account. In this paper, Gaussian process regression (GPR) is adapted for traffic modeling and prediction. The connection between self-similarity as a traffic characteristic and GPR parameters has been driven and exerted to build of a new Hurst estimation method based on machine learning techniques. This led to propose self-similar covariance functions for enhancing prediction accuracy of GPR. The proposed GPR model has been applied for Hurst estimation as well as for traffic prediction on real traffic traces at different time-scales. The experimental results show the employment of self-similar covariance functions increases generalization ability of GPR for traffic modeling and prediction.

In this paper, we propose a novel and general framework for dimensionality reduction, called Relational Fisher Analysis (RFA). Unlike traditional dimensionality reduction methods, such as linear discriminant analysis (LDA) and marginal Fisher analysis (MFA), RFA seamlessly integrates relational information among data into the representation learning framework, which in general provides strong evidence for related data to belong to the same class. To address nonlinear dimensionality reduction problems, we extend RFA to its kernel version. Furthermore, the convergence of RFA is also proved in this paper. Extensive experiments on documents understanding and recognition, face recognition and other applications from the UCI machine learning repository demonstrate the effectiveness and efficiency of RFA.

In this letter, a very simple no-reference image quality assessment (NR-IQA) model for JPEG compressed images is proposed. The proposed metric called median of unique gradients (MUG) is based on the very simple facts of unique gradient magnitudes of JPEG compressed images. MUG is a parameterless metric and does not need training. Unlike other NR-IQAs, MUG is independent to block size and cropping. A more stable index called MUG$^+$is also introduced. The experimental results on six benchmark datasets of natural images and a benchmark dataset of synthetic images show that MUG is comparable to the state-of-the-art indices in the literature. In addition, its performance remains unchanged for the case of the cropped images in which block boundaries are not known. The MATLAB source code of the proposed metrics is available athttps://dl.dropboxusercontent.com/u/74505502/MUG.mandhttps://dl.dropboxusercontent.com/u/74505502/MUGplus.m.

The performance of handwriting recognition systems is dependent on the features extracted from the word image. A large body of features exists in the literature, but no method has yet been proposed to identify the most promising of these, other than a straightforward comparison based on the recognition rate. In this paper, we propose a framework for feature set evaluation based on a collaborative setting. We use a weighted vote combination of recurrent neural network (RNN) classifiers, each trained with a particular feature set. This combination is modeled in a probabilistic framework as a mixture model and two methods for weight estimation are described. The main contribution of this paper is to quantify the importance of feature sets through the combination weights, which reflect their strength and complementarity. We chose the RNN classifier because of its state-of-the-art performance. Also, we provide the first feature set benchmark for this classifier. We evaluated several feature sets on the IFN/ENIT and RIMES databases of Arabic and Latin script, respectively. The resulting combination model is competitive with state-of-the-art systems.

In this paper, we propose a new unsupervised change detection method designed to analyze multispectral remotely sensed image pairs. It is formulated as a segmentation problem to discriminate the changed class from the unchanged class in the difference images. The proposed method is in the category of the committee machine learning model that utilizes an ensemble of classifiers (i.e., the set of segmentation results obtained by several thresholding methods) with a dynamic structure type. More specifically, in order to obtain the final “change/no-change” output, the responses of several classifiers are combined by means of a mechanism that involves the input data (the difference image) under an iterative Bayesian-Markovian framework. The proposed method is evaluated and compared to previously published results using satellite imagery.

Cloud computing is increasingly changing the landscape of computing, however, one of the main issues that is refraining potential customers from adopting the cloud is the security. Network functions virtualization together with software-defined networking can be used to efficiently coordinate different network security functionality in the network. To squeeze the best out of network capabilities, there is need for algorithms for optimal placement of the security functionality in the cloud infrastructure. However, due to the large number of flows to be considered and complexity of interactions in these networks, the classical placement algorithms are not scalable. To address this issue, we elaborate an optimization framework, namely OCDO, that provides adequate and scalable network security provisioning and deployment in the cloud. Our approach is based on an innovative multistage approach that combines together decomposition and segmentation techniques to the problem of security functions placement while coping with the complexity and the scalability of such an optimization problem. We present the results of multiple scenarios to assess the efficiency and the adequacy of our framework. We also describe our prototype implementation of the framework integrated into an open source cloud framework, i.e. Open stack.

In this work, a new region-based, multipath-enabled packet routing is presented and called Smart Packet Routing. The proposed approach provides several opportunities to re-distribute the smartness and decision making among various elements of a network including the packets themselves toward providing a decentralized solution for SDNs. This would bring efficiency and scalability, and therefore also lower environmental footprint for the ever-growing networks. In particular, a region-based representation of the network topology is proposed which is then used to describe the routing actions along the possible paths for a packet flow. In addition to a region stack that expresses a partial or full regional path of a packet, QoS requirements of the packet (or its associated flow) is considered in the packet header in order to enable possible QoS-aware routing at region level without requiring a centralized controller.

The first competition on the MultiSpectral Text Extraction (MS-TEx) from historical document images has been organized in conjunction with the ICDAR 2015 conference. The goal of this contest is evaluation of the most recent advances in text extraction from historical document images captured by a multispectral imaging system. The MS-TEx 2015 dataset contains 10 handwritten and machine-printed historical document images along with eight spectral images for each image. This paper provides a report on the methodology and performance of the five submitted algorithms by various research groups across the world. The objective evaluation and ranking was performed by using well-known evaluation metrics of binarization and classification.

In this work, a multiple-expert binarization framework for multispectral images is proposed. The framework is based on a constrained subspace selection limited to the spectral bands combined with state-of-the-art gray-level binarization methods. The framework uses a binarization wrapper to enhance the performance of the gray-level binarization. Nonlinear preprocessing of the individual spectral bands is used to enhance the textual information. An evolutionary optimizer is considered to obtain the optimal and some suboptimal 3-band subspaces from which an ensemble of experts is then formed. The framework is applied to a ground truth multispectral dataset with promising results. In addition, a generalization to the cross-validation approach is developed that not only evaluates generalizability of the framework, it also provides a practical instance of the selected experts that could be then applied to unseen inputs despite the small size of the given ground truth dataset.

In this paper, we present an exhaustive comparison of techniques for classification of signature data extracted from gyroscope and accelerometer devices. Since there exists large pool of classifiers and features for this kind of data, in order to provide a guide in choosing a particular setup, we decided to explore performance of these methods in a comparative study, which is a missing factor of current works on the topic. Also, we propose a framework for the combination of evaluated techniques in order to achieve a higher precision of the final classifier. The evaluated factors are: transformation of the time-series data into a fixed-size vector, classification methods and the performance of generative techniques without fixed-size input.

Coronary arteries (CA) segmentation from an angiographic sequence is essential to guide the cardiologists during percutaneous interventions for the treatment and diagnosis of pathologies. Segmentation of the CA from X-ray angiograms is a very challenging problem due to the changes in contrast in the sequence in addition to the CA's complex topology. In this paper, we propose a hierarchical segmentation method that extends the Vessel Walker model using a temporal prior and multiscale information to extract CA with a higher level of accuracy. In this method, the vessel located in frame Itat time t is extracted by utilizing the segmentation result at frame It−1together with Histogram of Oriented Gradient (HOG) features and a shape matching technique. Our experiments conducted on five paediatric angiograms have shown promising qualitative and quantitative results with a mean Dice coefficient of 64% and 53% Recall and 85% in Precision.

Traditional deblurring algorithms are often focused on natural-scaled images, which are not adapted for document texts and images without having some negative impacts on the accuracy of the OCR and the visual quality. In this paper, we propose a gradient histogram preservation method. An effective optimization method was developed and achieves satisfying results for kernel estimation. By combining the gradient histogram preservation prior with conventional image deblurring methods, it significantly improves the simulations and experimental results on document images and a high SSIM is achieved with the proposed method.

Multi-tenant flexible, scalable and autonomic virtual networks isolation has long been a goal of the network research and industrial community. For today's Software-Defined Networking (SDN) platforms, providing cloud tenants requirements for scalability, elasticity, and transparency is far from straightforward. SDN programmers typically enforce strict and inflexible traffic isolation resorting to low-level encapsulations mechanisms which help and facilitate network programmer reasoning about their complex slices behavior. In this paper, we propose SD-NMS, a novel software-defined architecture overcoming SDN and encapsulation techniques limitations. SD-NMS lifts several network virtualization roadblocks by combining these two separate approaches into an unified design. SD-NMS design leverages the benefits of SDN to provide Layer 2 (L2) isolation coupled with network overlay protocols with simple and flexible virtual tenant slices abstractions. This yields a network virtualization architecture that is both flexible, scalable and secure on one side, and self-manageable on the other. The experiment results showed that the proposed design offers negligible overhead and guarantees the network performance while achieving the desired isolation goals.

Traditional enterprise network security is based on the deployment of security appliances placed on some specific locations filtering, monitoring the traffic going through them. In this perspective, security appliances are chained in specific order to perform different security functions on the traffic. In the cloud, the same approach is often adopted using virtual security appliances to protect traffic for different virtual applications with the challenge of dealing with the flexible and elastic nature of the cloud. In this paper, we investigate the problem of placing virtual security appliances within the data center in order to minimize network latency and computing costs for security functions while maintaining the required sequential order of traversing virtual security appliances. We propose a new algorithm computing the best place to deploy these virtual security appliances in the data center. We further integrated our placement algorithm in an open source cloud framework, i.e. Openstack, in our test laboratory. The preliminary results show that we are placing the virtual security appliances in the required sequential order while improving the efficiency compared to the current default placement algorithm in Openstack.

Cloud computing has a central role to play in meeting today׳s business requirements. However, Distributed Denial-of-Service (DDoS) attacks can threaten the availability of cloud functionalities. In recent years, many effort has been expended to detect the various DDoS attack types. In this survey paper, our concentration is on how to mitigate these attacks. We believe that cloud computing technology can substantially change the way we respond to a DDoS attack, based on a number of new characteristics, which were introduced with the advent of this technology. We first present a new taxonomy of DDoS mitigation strategies to organize the work. Then, we go on to discuss the main features of existing DDoS mitigation strategies and explain their functionalities in the cloud environment. Afterwards, we show how the existing DDoS mechanisms fit into the network topology of the cloud. Finally, we discuss some of these DDoS mechanisms in detail, and compare their behavior in the cloud. Our objective is to show how these characteristics bring a novel perspective to existing DDoS mechanisms, and so give researchers new insights into how to mitigate DDoS attacks in the cloud computing.

In this work, based on the local phase information of images, an objective index, called the feature similarity index for tone-mapped images (FSITM), is proposed. To evaluate a tone mapping operator (TMO), the proposed index compares the locally weighted mean phase angle map of an original high dynamic range (HDR) to that of its associated tone-mapped image calculated using the output of the TMO method. In experiments on two standard databases, it is shown that the proposed FSITM method outperforms the state-of-the-art index, the tone mapped quality index (TMQI). In addition, a higher performance is obtained by combining the FSITM and TMQI indices.

This paper presents a novel preprocessing method of color-to-gray document image conversion. In contrast to the conventional methods designed for natural images that aim to preserve the contrast between different classes in the converted gray image, the proposed conversion method reduces as much as possible the contrast (i.e., intensity variance) within the text class. It is based on learning a linear filter from a predefined data set of text and background pixels that: 1) when applied to background pixels, minimizes the output response and 2) when applied to text pixels, maximizes the output response, while minimizing the intensity variance within the text class. Our proposed method (called learning-based color-to-gray) is conceived to be used as preprocessing for document image binarization. A data set of 46 historical document images is created and used to evaluate subjectively and objectively the proposed method. The method demonstrates drastically its effectiveness and impact on the performance of state-of-the-art binarization methods. Four other Web-based image data sets are created to evaluate the scalability of the proposed method.

Environmental footprint resulting from datacenters activities can be reduced by both energy efficiency and renewable energy in a complementary fashion thanks to cloud computing paradigms. In a cloud hosting multi-tenant applications, virtual service providers can be provided with real-time recommendation techniques to allocate their virtual resources in edge, core, or access layers in an optimal way to minimize costs and footprint. Such a dynamic technique requires a flexible and optimized networking scheme to enable elastic virtual tenants spanning multiple physical nodes. In this paper, we investigate an environment-aware paradigm for virtual slices that allows improving energy efficiency and dealing with intermittent renewable power sources. A virtual slice consists of optimal flows assigned to virtual machines (VMs) in a virtual data center taking into account traffic requirements, VM locations, physical network capacity, and renewable energy availability. Considering various cloud consolidation schemes, we formulate and then propose an optimal solution for virtual slice assignment problem. Simulations on the GSN showed that the proposed model achieves better performance than the existing methods with respect to network footprint reductions.

Scalability for distributed Data Center Networks (DCNs) has long been a goal of the network research and industrial community. To support dynamically increasing demands from multi-tenants, the network providers have to duplicate or share virtual resources for satisfying tenants' requests. However, current Software-Defined Networking (SDN) architectures have major drawbacks including lack of scalability and cross Virtual Tenant Network (VTN) communication. They rely only on the flexibility of control plane and neglect management plane important role. SDN scalability bottleneck affects directly the network/VTN scalability. In front of the fast growing network, it is widely accepted that the network of the future will require more capabilities such as self-awareness, self-control and self-management. At the core of these challenges is providing elastic isolation for multi-tenancy and involving tenant in management and control to reach the scalability objective and reduce the complexity of management operations of large DCNs. To address these challenges, the Open virtual Network Management and Security (Open vNMS) is proposed for supporting transparent multi-tenancy while both network and VTN scalability is solved. Basing on elastic L2 isolation using SDN components' flexibility, we design an autonomic architecture to provide self-control, self-management and self-adaptive capabilities for the network. The experiment results showed that the proposed design offers negligible overhead and guarantees the network performance.

Cloud computing has recently emerged compelling paradigm by introducing several characteristics such as on-demand self-service, broad network access, resource pooling, rapid elasticity, and measured service. Despite the fact that cloud computing offers huge cost benefits for companies, the unique security challenges have been introduced in a cloud environment that make risk assessment challenging. Cloud consumers need a protection to their cloud applications against cyber attacks. Although some security controls and policies are devised for each element of cloud computing, we need a framework with overall quantitative risk assessment model. The aim of this paper is to propose a framework for assessing the security risks associated with cloud computing platforms. The fully quantitative, iterative, and incremental approach enables cloud customer/provider to assess and manage cloud security risks. A proper result of risk assessment leads to have appropriate risk management mechanism for mitigating risks and reach to an acceptance security level.

Old document images often suffer from different types of degradation that render their binarization a challenging task. In this paper, a new binarization algorithm for degraded document images is presented. The method is based on active contours evolving according to intrinsic geometric measures of the document image, Nib lack's thresholding is also used to control the active contours propagation. The validity of the proposed method is demonstrated on both recent and historical document images including different types of degradations, the results are compared with a number of known techniques in the literature.

This paper presents a novel verification approach towards improvement of handwriting recognition systems using a word hypotheses rescoring scheme by Deep Belief Networks (DBNs). A recurrent neural network based sequential text recognition system is used at first to provide the N-best recognition hypotheses of word images. Word hypotheses are aligned with the word image to obtain the character boundaries. Then, a verification approach using a DBN classifier is performed for each character segments. DBNs are recently proved to be very effective for a variety of machine learning problems. The character probabilities obtained from DBNs are next combined with the base recognition system. Finally, the N-best recognition hypotheses list is reranked according to the new score. We have compared our proposed approach with an MLP based rescoring approach on the Rimes dataset. The results obtained show that the verification approach using DBNs outperforms that of MLP systems.

Most of the classical methods for degraded document binarization are based on the pixel gray level intensity or on simple pixel neighborhood information such as mean or variance to compute the binarization threshold. Moreover, these information are extracted from the spatial domain of the document image which are not very discriminative. In this paper, we propose to estimate texture information based on Gabor filters for ancient degraded documents. First, the dominant slant angle of the document image script is computed by using the Fourier transform. Then, this dominant angle is used within a weighted sum of angles in a Gabor filter bank in order to capture more efficiently the document image foreground (text). This information, combined with the variance and the mean extracted respectively from spatial and frequency domains are used for estimating the binarization threshold. Three variants are used for evaluating the performance of Gabor filter bank, which are based on Niblack's, Sauvola's, and Wolf's thresholds. Experimental results conducted on DIBCO Datasets show that the proposed method is more appropriate for poor contrasted documents and ink-bleed through degradations.

In this paper we present an efficient face verification system based on the fusion of multi-scale multi-descriptor local binary features. First, the face is divided into regions and each region is divided into several patches. For each patch and at every specific scale, the statistics of the baseline Local Binary Pattern (LBP), the Local Phase Quantization (LPQ) and the recently proposed Binarized Statistical Image Feature (BSIF) are summarized by histograms. The histograms of different patches belonging to the same region are concatenated to form a highly dimensional feature vector representing a specific descriptor at a specific scale. Second, we propose an efficient dimensionality reduction technique based on Exponential Linear Discriminant Analysis EDA coupled with Within-Class Covariance Normalization (WCCN) to downgrade the effect of the directions of high intravariability and to enhance the discrimination power of the EDA. The projected histograms for each region are scored using the cosine similarity metric. Lastly, the different region scores corresponding to different descriptors at different scales are fused using support vector machine classifier (SVM). Experimental verification results demonstrate that the proposed authentication pipeline outperforms all the existing systems on the XM2VTS controlled database and interestingly compete with the top performing systems on the challenging LFW database.

This article deals with a serious form of degradation that often affects the readability of historical document images: the invisibility of text or ink. Due to wear over long periods of storage, the ink may become invisible to the human eye, an undesirable situation for scholars (i.e. Indian Ocean World project (IOW1, with whom we are working closely). Because only the class of ink is known a priori (reference), it can be considered as a target to be detected. This can be achieved by designing a linear filter that maximizes an energy function while minimizing the false detection of document image background elements. For each document image in which the ink is targeted, an internal reference is defined by a new self-referencing strategy. The proposed method is compared with a state-of-the-art methods, and validated on samples of real historical document images.

We present a supervised model for tensor dimensionality reduction, which is called large margin low rank tensor analysis (LMLRTA). In contrast to traditional vector representation-based dimensionality reduction methods, LMLRTA can take any order of tensors as input. And unlike previous tensor dimensionality reduction methods, which can learn only the low-dimensional embeddings with a priori specified dimensionality, LMLRTA can automatically and jointly learn the dimensionality and the low-dimensional representations from data. Moreover, LMLRTA delivers low rank projection matrices, while it encourages data of the same class to be close and of different classes to be separated by a large margin of distance in the low-dimensional tensor space. LMLRTA can be optimized using an iterative fixed-point continuation algorithm, which is guaranteed to converge to a local optimal solution of the optimization problem. We evaluate LMLRTA on an object recognition application, where the data are represented as 2D tensors, and a face recognition application, where the data are represented as 3D tensors. Experimental results show the superiority of LMLRTA over state-of-the-art approaches.

In this paper, a phase-based binarization model for ancient document images is proposed, as well as a postprocessing method that can improve any binarization method and a ground truth generation tool. Three feature maps derived from the phase information of an input document image constitute the core of this binarization model. These features are the maximum moment of phase congruency covariance, a locally weighted mean phase angle, and a phase preserved denoised image. The proposed model consists of three standard steps: 1) preprocessing; 2) main binarization; and 3) postprocessing. In the preprocessing and main binarization steps, the features used are mainly phase derived, while in the postprocessing step, specialized adaptive Gaussian and median filters are considered. One of the outputs of the binarization step, which shows high recall performance, is used in a proposed postprocessing method to improve the performance of other binarization methodologies. Finally, we develop a ground truth generation tool, called PhaseGT, to simplify and speed up the ground truth generation process for ancient document images. The comprehensive experimental results on the DIBCO'09, H-DIBCO'10, DIBCO'11, H-DIBCO'12, DIBCO'13, PHIBD'12, and BICKLEY DIARY data sets show the robustness of the proposed binarization method on various types of degradation and document images.

In handwriting recognition, design of relevant feature is a very important but daunting task. On one hand, handcraft design of features is difficult, depending on expert knowledge and on heuristics. On the other hand, biologically inspired neural networks are able to learn automatically features from the input image, but requires a good underlying model. The goal of this paper is to evaluate the performance of automatically learned features compared to handcrafted features, as they provide a promising alternative to the difficult task of features handcrafting. In this work, the recognition model is based on the long short-term memory (LSTM) and connectionist temporal classification (CTC) neural networks. This model has been shown to outperform the well-known HMM model for various handwriting tasks, thanks to its reliable probabilistic modeling. In its multidimensional form, called MDLSTM, this network is able to automatically learn features from the input image. For evaluation, we compare the MDLSTM learned features and four state-of-the-art handcrafted features. The IFN/ENIT database has been used as benchmark for Arabic word recognition, where the results are promising.

Human ground-truthing is the manual labelling of samples (pixels for example) to generate reference data without any automatic algorithm help. Although a manual ground-truth is more accurate than a machine ground-truth, it still suffers from mislabeling and/or judgement errors. In this paper we propose a new method of ground-truth estimation using multispectral (MS) imaging representation space for the sake of document image binarization. Starting from the initial manual ground-truth, the proposed classification method aims to select automatically some samples with correct labels (well-labeled pixels) from each class for the training phase, then reassign new labels to the document image pixels. The classification scheme is based on the cooperation of multiple classifiers under some constraints. A real data set of MS historical document images and their ground-truth is created to demonstrate the effectiveness of the proposed method of ground-truth estimation.

In recent years, a large number of binarization methods have been developed, with varying performance generalization and strength against different benchmarks. In this work, to leverage on these methods, an ensemble of experts (EoE) framework is introduced, to efficiently combine the outputs of various methods. The proposed framework offers a new selection process of the binarization methods, which are actually the experts in the ensemble, by introducing three concepts: confident ness, endorsement and schools of experts. The framework, which is highly objective, is built based on two general principles: (i) consolidation of saturated opinions and (ii) identification of schools of experts. After building the endorsement graph of the ensemble for an input document image based on the confident ness of the experts, the saturated opinions are consolidated, and then the schools of experts are identified by thresholding the consolidated endorsement graph. A variation of the framework, in which no selection is made, is also introduced that combines the outputs of all experts using endorsement-dependent weights. The EoE framework is evaluated on the set of participating methods in the H-DIBCO'12 contest and also on an ensemble generated from various instances of grid-based Sauvola method with promising performance.

For the purpose of facilitating benchmark contributions for binarization methods, a new fast ground truthing approach, called the PhaseGT, is proposed. This approach is used for building the 1stgroundtruthed Persian Heritage Image Binarization Dataset (PHIBD 2012). The PhaseGT is a semiautomatic approach to ground truthing of images of any language, especially designed for historical document images. The main goal of the PhaseGT is to accelerate the ground truthing process and reduce the manual ground truthing effort. It uses the phase congruency features to preprocess the input image and to provide a more accurate initial binarization to the human expert who performs the manual part. This preprocessing is in turn based on a priori knowledge that is provided by human user. The PHIBD 2012 dataset contains 15 historical document images with their corresponding ground truth binary images. The historical images in the dataset suffer from various types of degradation. It has been also divided into two subsets of training and testing images for those binarization methods that use learning approaches.

Preprocessing and post processing steps significantly improve the performance of binarization methods, especially in the case of severely-degraded historical documents. In this paper, an unsupervised post processing method is introduced based on the phase-preserved denoised image and also phase congruency features extracted from the input image. The core of the method consists of two robust mask images that can be used to cross out false positive pixels on the output of the binarization method. First, a mask with a high recall value is obtained from the denoised image using morphological operations. In parallel, a second mask is obtained based on phase congruency features. Then, a median filter is used to remove noise on these two masks, which then are used to correct the output of any binarization method. This approach was tested along with several state-of the-art binarization methods on the DIBCO'09, H-DIBCO'10, DIBCO'11 and H-DIBCO'12 datasets with promising and robust results. Furthermore, the high performance of the proposed masks shows their potential use as unsupervised semi-ground truth generator for learning-based binarization methods.

Incremental learning, especially when learning from a scratch, has a lot of interest for online gesture recognition. However the lack of learning examplers combined to low computational cost suggests building robust and efficient learning machines. In this paper we introduce a hybrid model of ART-2A neural network combined to Takagi-Sugeno (TS) neuro-fuzzy network. The latter model is applied for online handwritten gesture recognition, when the learning is starting from scratch and no class information, such as gesture type or number of classes, is predefined. Moreover, using ART-2A neural network and our novel distance measure, the computational complexity of the whole model decreases while preserving high accuracy. Furthermore, we exploit the forgetting dilemma of online learning by introducing a competitive Recursive Least Squares method for TS models. Together, all the modeling has shown promising results.

In this paper we present a novel approach towards retrieval of documents with pictorial data. Many prior works focused on word-spotting and text-based retrieval, but none of these techniques handled the retrieval of pictorial part of documents. In this paper, we present a new method that allows users to retrieve any visual data from documents, based on a sketch example. It mainly emphasizes all three main aspects of a visual retrieval system: feature representation, indexing and retrieval. Especially we focus on the design of salient descriptors, capable of capturing unique mapping from sketched images to document illustrations. We evaluate several approaches towards feature representation and indexing, with the aim to maximize the performance of our method. Our proposed technique is highly useful to complement word-spotting technique, when the indexed documents are composed of mixture of visual and textual data. This technique has shown promising results, both on pictorial data automatically extracted from documents as well for those selected by users as regions of interest.

In the literature, many dimensionality reduction methods have been proposed and applied to recognition tasks, including handwritten digits recognition, character recognition and string recognition. However, it is usually difficult for the researchers to decide which method is the optimal choice for the problem at hand. In this paper, we empirically compare some supervised dimensionality reduction methods on handwritten digits recognition, English letter recognition and ancient Arabic sub word recognition, to evaluate their performance on the recognition tasks. These compared methods include traditional linear dimensionality reduction approach (linear discriminant analysis, LDA), locality-based manifold learning approach (marginal Fisher analysis, MFA) and relational learning approach (probabilistic relational principal component analysis, PRPCA). Experimental results and statistical tests show that locality-based manifold learning approach (MFA) generally performs well in terms of recognition accuracy, but with high computational complexity, traditional linear dimensionality reduction approach (LDA) is efficient, but not necessarily to deliver the best result, relational learning approach (PRPCA) is promising, and more efforts should be dedicated to this area.

Curvilinear structure detection filters are crucial building blocks in many medical image processing applications, where they are used to detect important structures, such as blood vessels, airways, and other similar fibrous tissues. Unfortunately, most of these filters are plagued by an implicit single structure direction assumption, which results in a loss of signal around bifurcations. This peculiarity limits the performance of all subsequent processes, such as understanding angiography acquisitions, computing an accurate segmentation or tractography, or automatically classifying image voxels. This paper presents a new 3-D curvilinear structure detection filter based on the analysis of the structure ball, a geometric construction representing second order differences sampled in many directions. The structure ball is defined formally, and its computation on a discreet image is discussed. A contrast invariant diffusion index easing voxel analysis and visualization is also introduced, and different structure ball shape descriptors are proposed. A new curvilinear structure detection filter is defined based on the shape descriptors that best characterize curvilinear structures. The new filter produces a vesselness measure that is robust to the presence of X- and Y-junctions along the structure by going beyond the single direction assumption. At the same time, it stays conceptually simple and deterministic, and allows for an intuitive representation of the structure's principal directions. Sample results are provided for synthetic images and for two medical imaging modalities.

With raising concerns about global warming and environmental impacts of Greenhouse Gases (GhGs) emissions, energy efficiency and carbon footprint reduction attracted many researchers to provide efficient models and tools for energy, carbon, and cost estimation and management. In this paper, a model for measuring the energy consumption and carbon footprint of an individual virtual machine is presented based on resource usage and performance monitoring counters. A simple cost model is represented in order to evaluate the energy consumption and carbon footprint models. The model evaluated on a simulated virtual private cloud with different methodologies such as server consolidation and multi-level grouping heuristic algorithms. The results show that such heuristic algorithms are able to significantly reduce the cost of energy and carbon footprint of an individual virtual machine in comparison with other methodologies such as server consolidation. The results also show that this cost reduction efficiency is positively correlated to the increase in carbon footprint tax rates.

Perceptual analysis is an interesting topic in the field of image processing, and can be considered a missing link between image processing and human vision. Of the various forms of perception, one of the most important and best known is shape perception. In this work, a framework based on the online non local patch means (NLPM) method is developed, which is designed to infer possible perceptual observations of an input image using the knowledge images provided. Thanks to the speed of online NLPM, the proposed method can simulate the transformation of the input image to the final perceptual image in real time. In order to improve the performance of the method, a hidden chain series is considered for the model that delivers faster convergence. The capability of the method is evaluated on several well-known perceptual examples.

A 2D/3D nonrigid registration method is proposed that brings a 3D centerline model of the coronary arteries into correspondence with bi-plane fluoroscopic angiograms. The registered model is overlaid on top of interventional angiograms to provide surgical assistance during image-guided chronic total occlusion procedures, thereby reducing the uncertainty inherent in 2D interventional images. The proposed methodology is divided into two parts: global structural alignment and local nonrigid registration. In both cases, vessel centerlines are automatically extracted from the 2D fluoroscopic images, and serve as the basis for the alignment and registration algorithms. In the first part, an energy minimization method is used to estimate a global affine transformation that aligns the centerline with the angiograms. The performance of nine general purpose optimizers has been assessed for this problem, and detailed results are presented. In the second part, a fully nonrigid registration method is proposed and used to compensate for any local shape discrepancy. This method is based on a variational framework, and uses a simultaneous matching and reconstruction process to compute a nonrigid registration. With a typical run time of less than 3 s, the algorithms are fast enough for interactive applications. Experiments on five different subjects are presented and show promising results.

Data center energy efficiency and carbon footprint reduction have attracted a great deal of attention across the world for some years now, and recently more than ever. Live Virtual Machine (VM) migration is a prominent solution for achieving server consolidation in Local Area Network (LAN) environments. With the introduction of live Wide Area Network (WAN) VM migration, however, the challenge of energy efficiency extends from a single data center to a network of data centers. In this paper, intelligent live migration of VMs within a WAN is used as a reallocation tool to minimize the overall carbon footprint of the network. We provide a formulation to calculate carbon footprint and energy consumption for the whole network and its components, which will be helpful for customers of a provider of cleaner energy cloud services. Simulation results show that using the proposed Genetic Algorithm (GA)-based method for live VM migration can significantly reduce the carbon footprint of a cloud network compared to the consolidation of individual data center servers. In addition, the WAN data center consolidation results show that an optimum solution for carbon reduction is not necessarily optimal for energy consumption, and vice versa. Also, the simulation platform was tested under heavy and light VM loads, the results showing the levels of improvement in carbon reduction under different loads.

This paper presents two new time-frequency distributions based on kernels with compact support (KCS) namely the separable CB (SCB) and the polynomial CB (PCB) TFDs. The implementation of these distributions follows the method developed for the Cheriet-Belouchrani CB TFD. The performance of this family of TFDs is compared to the most known quadratic distributions through tests on multi-component signals with linear and nonlinear frequency modulations (FMs) considering the noise effects as well. Comparisons are based on the evaluation of an objective criterion namely the Boashash-Sucic's normalized instantaneous resolution performance measure that allows to provide the optimized TFD using a specific methodology. In all presented examples, the KCS TFDs have been shown to have a significant interference mitigation, with the component energy concentration around their respective instantaneous frequency laws being well preserved giving high resolution measure values.

The extraction and analysis of useful information from old document images is very important into cultural heritage preservation. In advanced research, where the goal is to separate the foreground (in general, text) from the background, image restoration and pattern classification techniques are used. Most of these methods consist of classifying the pixels based on their gray-scale value. In this paper, we propose to perform foreground pattern extraction using regions-of-interest (ROI) analysis and a maximum likelihood classifier designed for multispectral document images. As contribution, a new feature vector is proposed to improve discrimination between patterns that is embedded in a simple statistical classification method. The results, which are promising, are compared to the state-of-the-art.

In the context of handwriting recognition, word confusion networks (WCN) are convenient representations of alternative recognition candidates. They provide alignment for mutually exclusive words along with the posterior probability of each word. In this paper, we present a method for indexing on-line handwriting based on WCN. The proposed method exploits the information provided by WCN in order to enhance relevant keyword extraction. In addition, querying the index for a given keyword has worst case complexity O(log n), as compared to usual keyword spotting algorithms which run in O(n). Experiments show promising results in keyword retrieval effectiveness by using WCN when compared to keyword search over 1-best recognition results.

Bayesian reasoning provides an ideal basis for representing and manipulating uncertain knowledge, with the result that many interesting algorithms in machine learning are based on Bayesian inference. In this paper, we use the Bayesian approach with one and two levels of inference to model the semisupervised learning problem and give its application to the successful kernel classifier support vector machine (SVM) and its variant least-squares SVM (LS-SVM). Taking advantage of Bayesian interpretation of LS-SVM, we develop a semisupervised learning algorithm for Bayesian LS-SVM using our approach based on two levels of inference. Experimental results on both artificial and real pattern recognition problems show the utility of our method.

This paper describes the steps that have been undertaken in order to develop the IBN SINA database, which is designed to apply learning techniques in the processing and understanding of document images. The description of the preparation process, including preprocessing, feature extraction and labeling, is provided. The database has been evaluated using classification techniques, such as the SVM classifiers. In order to make the database compatible with these classifiers, the labels of the shapes have been translated into a set of bi-class problems. Promising results with the SVM classifiers have been obtained.

In this paper, we present a new method to design customizable self-evolving fuzzy rule-based classifiers. The presented approach combines an incremental clustering algorithm with a fuzzy adaptation method in order to learn and maintain the model. We use this method to build an evolving handwritten gesture recognition system, that can be integrated into an application to provide personalization capabilities. Experiments on an on-line gesture database were performed by considering various user personalization scenarios. The experiments show that the proposed evolving gesture recognition system continuously adapts and evolve according to new data of learned classes, and remains robust when introducing new unseen classes, at any moment during the lifelong learning process.

In this paper, we present a new method to design customizable self-evolving fuzzy rule-based classifiers. The presented approach combines an incremental clustering algorithm with a fuzzy adaptation method in order to learn and maintain the model. We use this method to build an evolving handwritten gesture recognition system. The self-adaptive nature of this system allows it to start its learning process with few learning data, to continuously adapt and evolve according to any new data, and to remain robust when introducing a new unseen class at any moment in the life-long learning process.

A new level-set based active contour method for the segmentation of small blood vessels and other elongated structures is presented. Its main particularity is the presence of a length increasing force in the contour driving equation. The effect of this force is to push the active contour in the direction of thin elongated shapes. Although the proposed force is not stable in general, our experiments show that with few precautions it can successfully be integrated in a practical segmentation scheme and that it helps to segment a longer part of the structures of interest. For the segmentation of blood vessels, this may reduce the amount of user interactivity needed: only a small region inside the structure of interest need to be specified.

The least squares support vector machine (LS-SVM) is an interesting variant of the SVM. It performs structural risk through margin-maximization and has excellent power of generalization. For some applications, it is more interesting to use the weighted LS-SVM where the impact of each training sample is controlled by weighting factors. In this paper, we consider the use of the weighted LS-SVM in semi-supervised learning. We propose an algorithm to perform this type of learning by extending the transductive SVM idea. We tested our algorithm on both artificial and real problems and demonstrate its usefulness comparing with other semi-supervised learning methods.

The goal of this paper is to correct bleed-through in degraded documents using a variational approach. The variational model is adapted using an estimated background according to the availability of the verso side of the document image. Furthermore, for the latter case, a more advanced model based on a global control, the flow field, is introduced. The solution of each resulting model is obtained using wavelet shrinkage or a time-stepping scheme, depending on the complexity and nonlinearity of the models. When both sides of the document are available, the proposed model uses the reverse diffusion process for the enhancement of double-sided document images. The results of experiments with real and synthesized samples are promising. The proposed model, which is robust with respect to noise and complex background, can also be applied to other fields of image processing.

A complete system for preprocessing and word spotting of very old historical document images is presented. Document images are processed for extraction of salient information using a word spotting technique which does not need line and word segmentation and is language independent.A multi-class library of connected components of document text is created based on six features. The spotting is performed using Euclidean distance measure enhanced by rotation and dynamic time wrapping transforms. The method is applied to a dataset from Juma Al Majid Center (Dubai)with promising results. A promising performance of the word spotting technique is obtained using an automatic preprocessing stage. In this stage, using content-level classifiers, accurate stroke pixels are extracted in a robust way. The preprocessed document images are also more legible to the end user and are less costly to archive and transfer.

Segmentation of ancient documents is challenging. In the worst cases, text characters become fragmented as the results of strong degradation processes. New active contour methods allow to handle difficult cases in a spatially coherent fashion. However, most of those method use a restrictive, a priori shape information that limit their application. In this work, we propose to address this issue by combining two complementary approaches. First, multi-level classifiers, which take advantage of the stroke width a priori information, allow to locate candidate character pixels. Second, a level set active contour scheme is used to identify the boundary of a character. Tests have been conducted on a set of ancient degraded Hebraic character images. Numerical results are promising.

A novel method for the segmentation of double-sided ancient document images suffering from bleed-through effect is presented. It takes advantage of the level set framework to provide a completely integrated process for the segmentation of the text along with the removal of the bleed-through interfering patterns. This process is driven by three forces: 1) a binarization force based on an adaptive global threshold is used to identify region of low intensity, 2) a reverse diffusion force allows for the separation of interfering patterns from the true text, and 3) a small regularization force favors smooth boundaries. This integrated method achieves high quality results at reasonable computational cost, and can easily host other concepts to enhance its performance. The method is successfully applied to real and synthesized degraded document images. Also, the registration problem of the double-sided document images is addressed by introducing a level set method; the results are promising.

In this paper we investigate the use of linguistic information given by language models to deal with word recognition errors on handwritten sentences. We focus especially on errors due to out-of-vocabulary (OOV) words. First, word posterior probabilities are computed and used to detect error hypotheses on output sentences. An SVM classifier allows these errors to be categorized according to defined types. Then, a post-processing step is performed using a language model based on part-of-speech (POS) tags which is combined to the n-gram model previously used. Thus, error hypotheses can be further recognized and POS tags can be assigned to the OOV words. Experiments on on-line handwritten sentences show that the proposed approach allows a significant reduction of the word error rate.

This paper presents a novel approach for skew correction of documents. Skew correction is modeled as an optimization problem, and for the first time, particle swarm optimization (PSO) is used to solve skew optimization. A new objective function based on local minima and maxima of projection profiles is defined, and PSO is utilized to find the best angle that maximizes differences between values of local minima and maxima. In our approach, local minima and maxima converge to the locations of lines and spaces between lines. Results of our skew correction algorithm are shown on documents written in different scripts such as Latin and Arabic related scripts (e.g. Arabic, Farsi,Urdu,...). Experiments show that our algorithm can handle a wide range of skew angles, also it is robust to gray level and binary images of different scripts.

The non-local means filter is one of powerful denoising methods which allows participation of far, but proper pixels in the denoising process. Although the weights of non-similar pixels are very small, high number of these pixels results in introduction of blur. In this work, we introduce an automatic and robust method to select the best candidate pixels based on their similarity to the target pixel. This method is based on graphs partitioning and uses Markovian clustering on the pixel adjacency graph (PAG). In this way, a set of relevant pixels is obtained that is used in weighted averaging for denoising each pixel. To evaluate the method, denoising of the natural images is conducted, and the results are compared to the standard NLM filter and the SVD-based method. The results are promising.

Help-training for semi-supervised learning was proposed in our previous work in order to reinforce self-training strategy by using a generative classifier along with the main discriminative classifier. This paper extends the Help-training method to least squares support vector machine (LS-SVM) where labeled and unlabeled data are used for training. Experimental results on both artificial and real problems show its usefulness when comparing with other classical semisupervised methods.

The least squares support vector machine (LS-SVM), like the SVM, is based on the margin-maximization performing structural risk and has excellent power of generalization. In this paper, we consider its use in semisupervised learning. We propose two algorithms to perform this task deduced from the transductive SVM idea. Algorithm 1 is based on combinatorial search guided by certain heuristics while Algorithm 2 iteratively builds the decision function by adding one unlabeled sample at the time. In term of complexity, Algorithm 1 is faster but Algorithm 2 yields a classifier with a better generalization capacity with only a few labeled data available. Our proposed algorithms are tested in several benchmarks and give encouraging results, confirming our approach.

In this paper, we propose to reinforce self-training strategy by using a generative classifier that may help the main discriminative classifier training in semi-supervised mode to label the unlabeled data. We called this semi-supervised strategy: help-training. We apply this method for training support vector machine with labeled and unlabeled data. Experimental results on both artificial and real problems show its usefulness comparing with other classical semi-supervised methods.

This study investigates Bayes classification of online Arabic characters represented by histograms of tangent differences and Gibbs modeling of the class-conditional probability density functions. The parameters of these Gibbs density functions are estimated following the Zhu, Wu, and Mumford constrained maximum entropy formalism, originally introduced for image and shape synthesis. We investigate two partition function estimation methods: one uses the training sample and the other draws from a reference distribution. The efficiency of the corresponding Bayes decision methods, and of a combination of these, is shown in experiments using a database of 9504 freely written samples by 22 scriptors. Comparisons to the nearest neighbor rule method and Kohonen neural network methods are provided.

This article describes an off-line handwritten Arabic words recognition system. Both explicit grapheme segmentation and feature extraction are originally designed for Latin cursive handwriting. The recognizer itself is a hybrid HMM/NN. We introduce a new shape-based alphabet for handwriting Arabic recognition which is intended to benefit from some specificities of Arabic writing. We performed several experiments using IFN/ENIT benchmark database to validate our approach. Our recognizer performs as close as the state of the art recognition rate with 87%. The latter results are indeed very encouraging as many perspectives and improvements may be considered. Especially, the explicit processing of dots and diacritics, therefore making use of more prior knowledge of Arabic writing specificities.

Support vector machine (SVM) is an interesting classifier that has an excellent power of generalization. In this paper, we consider SVM in semi-supervised learning. We propose to use an additional criterion with the standard formulation of the transductive SVM for reinforcing the classifier regularization. Also, we use a genetic algorithm for optimizing the objective function, since the transductive SVM yields a non-convex problem. We tested our algorithm on artificial and real data, which gives promising results in comparison with Joachims' algorithm known as SVMlight TSVM.

In a scale-space framework, the Gaussian kernel has some properties that make it unique. However, because of its infinite support, exact implementation of this kernel is not possible. To avoid this drawback, there exist two different approaches: approximating the Gaussian kernel by a finite support kernel, or defining new kernels with properties closed to the Gaussian. In this paper, we propose a polynomial kernel family with compact support which overcomes the Gaussian practical drawbacks while preserving a large number of the useful Gaussian properties. The new kernels are not obtained by approximating the Gaussian, though they are derived from it. We show that, for a suitable choice of kernel parameters, this family provides an approximated solution of the diffusion equation and satisfies some other basic constraints of the linear scale-space theory. The construction and properties of the proposed kernel are described, and an application in which handwritten data are extracted from noisy document images is presented.

Model selection for support vector machines concerns the tuning of SVM hyperparameters as C controlling the amount of overlap and the kernel parameters. Several criteria developed for tuning the SVM hyperparameters, may not be differentiable w.r.t. C, consequently, gradient-based optimization methods are not applicable. In this paper, we propose a new formulation for SVM that makes possible to include the hyperparameter C in the definition of the kernel parameters. Then, tuning hyperparameters for SVM is equivalent to choosing the best values of kernel parameters. We tested this new formulation for model selection by using the criterion of empirical error, technique based on generalization error minimization through a validation set. The experiments on different benchmarks show promising results confirming our approach.

Noise and complex backgrounds often make the thresholding of degraded document images difficult. In this paper, we propose a new threshold selection method to handle severely degraded document images. First, multiscale image description is adopted to analyze the image edge. From this, edge pixel pair information is derived and recorded by a graylevel co-occurrence matrix. An appropriate threshold value is chosen by measuring the edge pixel pair co-occurrence matrix. The new method is tested with degraded document images. The experimental results show it is resistant to noise and complex backgrounds.

A new fast training algorithm for the multilayer perceptron (MLP) is proposed. This new algorithm is based on the optimization of a mixed least square (LS) and a least fourth (LF) criterion producing a modified form of the standard back propagation algorithm (SBP). To determine the updating rules in the hidden layers, an analogous back propagation strategy used in the conventional learning algorithms is developed. This permits the application of the learning procedure to all the layers. Experimental results on benchmark applications and a real medical problem are obtained which indicates significant reduction in the total number of iterations, the convergence time, and the generalization capacity when compared to those of the SBP algorithm.

Tuning SVM kernel parameters is a an important step for achieving a high-performing learning machine. The usual automatic methods used to tune these parameters require an inversion of the Gram-Schmidt matrix or a resolution of an extra quadratic programming problem. In the case of a large dataset these methods require the addition of huge amounts of memory and a long CPU time to the already significant resources used in the SVM training. In this paper, we propose a fast method based on an approximation of the gradient of the empirical error along with incremental learning, which reduces the resources required both in terms of processing time and of storage space.

In this paper, we propose a comparison of several post-processing methods for estimating multi-class probabilities with standard support vector machines. The different approaches have been tested on a real pattern recognition problem with a large number of training samples. The best results have been obtained by using a "one against air coupling strategy along with a softmax function optimized by minimizing the negative log-likelihood of the training data. Finally, the analysis of the error-reject tradeoff have shown that SVM allows to estimate probabilities more accurate than a classical MLP, which is indeed promising in the view of incorporated within pattern recognition system using probabilistic framework.

Classification of sequential data using discriminative models such as support vector machines is very hard due to the variable length of this type of data. On the other hand, generative models such as HMMs have become the standard tool for representing sequential data due to their efficiency. This paper proposes a general generative-discriminative framework that uses HMMs to map the variable length sequential data into a fixed size P-dimensional vector (likelihood score) that can be easily classified using any discriminative model. The preliminary experiments of the framework on the MNIST database for handwritten digits have achieved a better recognition rate of 98.02% than that of standard HMMs (94.19%).

The paper investigates a new representation of shape and its use in handwritten on-line character recognition. This representation is based on the empirical distribution of features such as tangents, and tangent differences at distant points along the character signal. Recognition is carried out by a Kohonen associative memory (also called Kohonen self organizing feature map), trained using this representation, and the Hellinger distance, which measures the distance between distributions. We report on extensive experiments that show the pertinence of the representation and the superior performance of the scheme.

We propose, in this paper, a new non-linear exponential adaptive bi-dimensional (2D) filter for image modeling. The filter coefficients are updated with the least mean square (LMS) algorithm. Furthermore, the proposed nonlinear model is used for texture modeling with a 2D auto-regressive (AR) adaptive model. The characterization efficiency of the proposed exponential model is compared with the 2D linear AR model updated with the LMS algorithm. The comparison criteria is based on the computation of a characterization rate using the ratio of "between-class" variances with respect to "within-class" variances of the estimated coefficients. Extensive experiments show that the exponential model coefficients give better results in texture discrimination than those of the linear model, even in a noisy context.

Extraction of pertinent data from noisy document images with complex backgrounds remains a challenging problem in character recognition applications. It depends on the quality of the character segmentation. Over the last few decades, mathematical tools have been developed for this purpose. Several authors show that the Gaussian kernel is unique and offers many beneficial properties. In their recent work Remaki and Cheriet proposed a new kernel family with compact supports (KCS) that achieved good performance with accurate information extraction and reducing drastically time processing with regard to the Gaussian kernel. In this paper, we focus in further improving its efficiency by proposing a new separable version which itself has a compact support. Experiments, on real life data, from noisy gray level images, show fast and high performance with accurate results of such a kernel. A practical comparison is established between results obtained by using the KCS and the SKCS operators. Our comparison is based on the information loss and the gain in time processing.

In this paper, we propose a generalized shock model for the enhancement and restoration of degraded gray-level character images. This model is a quasi-linear hyperbolic partial differential equation with discontinuous coefficients, based on a recently-published work. The model was extended to peeling character shapes with a uniform and centered width of value K. These processing tools may significantly improve the subsequent stages of the character image recognition task. Both models have been tested on a set of gray-level character images of varying quality from the CEDAR database, and experimental results have proved the efficiency and effectiveness of such modeling. 1.

The purpose of this study is to investigate a method for high performance on-line Arabic characters recognition. This method is based on the use of Kohonen maps and their corresponding confusion matrices which serve to prune them of error-causing nodes, and to combine them consequently. We use two Kohonen maps obtained using two distinct character representations, namely, Fourier descriptors and tangents extracted along the characters online signals. The two Kohonen maps are then combined using a majority vote decision rule that, for each character, favors the most reliable map. This combination, without adding in any significant way to the process complexity, affords a much better recognition rate.

We present a method for automatically selecting the best filter to treat poor quality printed documents using image quality assessment. We introduce five quality measures to obtain information about the quality of the images, and morphological filters to improve their quality. A training set of 370 images was used to develop the system. Experimental results on the test set show a significant improvement in the recognition rate from 73.24% using no filter at all to 93.09% after applying a filter that was automatically selected.

A new direction in machine learning area has emerged from Vapnik's theory in support vectors machine (SVM) and its applications on pattern recognition. In this paper we propose a new SVM kernel family, called KMOD (kernel with moderate decreasing) with distinctive properties that allow better discrimination in the feature space. The experiments that we carry out show its effectiveness on synthetic and large-scale data. We found KMOD performs better than RBF and exponential RBF kernels on the two-spiral problem. In addition, a digit recognition task was processed using the proposed kernel. The results show, at least, comparable performances to state of the art kernels.

In this paper, a model-based scheme for recognizing and beautifying online hand-drawn sketches of electric diagrams is presented. The system uses a structural and topological relations matching mechanism that allows scale, translation, rotation invariant recognition. A simple prototype was developed and preliminary experimental results show how this technique, although simple, is efficient in recognizing such sketches.

Although promising results on the combination of character recognizers have been reported recently, the combination strategies can not be readily applied to the recognition of character strings due to m-n correspondence problems caused by segmentation errors. In this paper, we propose a new paradigm of combining multiple string recognizers and contribute a generic framework for off-line combination. We designed and implemented a graph based off-line combination system, StrCombo, which has achieved a substantial improvement over any one of the individual recognizers in a real-life application. This open combination system provides the possibility of further improving the performance of string recognizers when new recognizers and combination rules are available.

Global gray-level thresholding techniques such as Otsu's method, and local gray-level thresholding techniques such as edge-based segmentation or the adaptive thresholding method are powerful in extracting character objects from simple or slowly varying backgrounds. However, they are found to be insufficient when the backgrounds include sharply varying contours or fonts in different sizes. A stroke-model is proposed to depict the local features of character objects as double-edges in a predefined size. This model enables us to detect thin connected components selectively, while ignoring relatively large backgrounds that appear complex. Meanwhile, since the stroke width restriction is fully factored in, the proposed technique can be used to extract characters in predefined font sizes. To process large volumes of documents efficiently, a hybrid method is proposed for character extraction from various backgrounds. Using the measurement of class separability to differentiate images with simple backgrounds from those with complex backgrounds, the hybrid method can process documents with different backgrounds by applying the appropriate methods. Experiments on extracting handwriting from a check image, as well as machine-printed characters from scene images demonstrate the effectiveness of the proposed model.

Multiscale representation is a methodology that is being used more and more when describing real-world structures. Scale-space representation is one formulation of multiscale representation that has received considerable interest in the literature because of its efficiency in several practical applications and the distinct properties of the Gaussian kernel that generate the scale space. Together, some of these properties make the Gaussian unique. Unfortunately, the Gaussian kernel has two practical limitations: information loss caused by the unavoidable Gaussian truncation and the prohibitive processing time due to the mask size. We propose a new kernel family derived from the Gaussian with compact supports that are able to recover the information loss while drastically reducing processing time. This family preserves a great part of the useful Gaussian properties without contradicting the uniqueness of the Gaussian kernel. The construction and analysis of the properties of the proposed kernels are presented in this paper. To assess the developed theory, an application of extracting handwritten data from noisy document images is presented, including a qualitative comparison between the results obtained by the Gaussian and the proposed kernels.

Presents a generalization of a new kernel family with compact support in scale space which we recently published (Vision Interface, pp. 445-52, May 1999). We have shown that the proposed kernels are able to recover the information loss when using the Gaussian kernel, while they drastically reduce the processing time. Furthermore, the generalized kernel family preserves all the properties of the previous one and it offers other important properties, like the behavior of the first and second derivatives which are guaranteed to be the same as the Gaussian kernel one at any scale space. The latter property plays an important role in image processing, as we show in this paper. The construction and some properties shown in the previous version are recalled and some of the new properties of the new version are proven. An application of extracting handwritten data from noisy bi-level document images is presented to point out the practical impact of the improved version of the proposed kernels.

This paper proposes a model-based character extraction method. We model a pixel belonging to a character as a double-edge, whose range is defined by the stroke width, and whose intensity is proportional to the local contrast. By extracting such double-edge feature at a predefined stroke width, sharply changing as well as slowly varying backgrounds can be eliminated. A set of morphological operators is designed to extract the double-edge feature at each pixel of the raw image, and the characters are extracted by thresholding these features. A goal directed evaluation of extraction of the courtesy amount from bank cheques reveals the advantages of the proposed method over other existing methods. Visual inspection of legal amount extraction shows further promise in extracting characters from complex backgrounds.

This paper presents a new methodology to extract visual data contained in noisy gray-level images such as mail envelopes. Since the intensity changes may occur over a large range of spatial scales in these and other like images, we adopt the multiscale approach to extract good quality data that might be used in further processing and recognition processes. We have already shown in a previous paper its effective use in full data extraction. In this paper, we will give an advanced formalism of this result, referred to as a top–down approach. Then, we will present a new and opposite approach, referred to as a bottom–up approach. The differences and characteristics of both approaches are highlighted. Experiments have been conducted on real life data from the data base provided by CEDAR at SUNY Buffalo, to assess the effectiveness of the proposed paradigm; this reveals its improved robustness and accuracy over the top–down approach; such a result might be useful for a wide range of applications in the field of image processing and enhancement.

In this correspondence, we present a general recursive approach for image segmentation by extending Otsu's (1978) method. The new approach has been implemented in the scope of document images, specifically real-life bank checks. This approach segments the brightest homogeneous object from a given image at each recursion, leaving only the darkest homogeneous object after the last recursion. The major steps of the new technique and the experimental results that illustrate the importance and the usefulness of the new approach for the specified class of document images of bank checks will be presented.

Presents a model for reading cursive scripts which has an architecture inspired by a reading model and which is based on perceptual concepts. We limit the scope of our study to the off-line recognition of isolated cursive words. First of all, we justify why we chose McClelland & Rumelhart's (1981) reading model as the inspiration for our system. A brief resume/spl acute/ of the method's behavior is presented and the main originalities of our model are underlined. After this, we focus on the new updates added to the original system: a new baseline extraction module, a new feature extraction module and a new generation, validation and hypothesis insertion process. After implementation of our method, new results have been obtained on real images from a training set of 184 images, and a testing set of 100 images, and are discussed. We are concentrating now on validating the model using a larger database.

A perfect segmentation method would be capable to segment words in letters. It would be then possible to define a process on letters. Unfortunately, such a method is almost impossible to obtain due to the nature of handwritten words. To tackle this problem, our approach segments the word into graphemes. We propose in this paper an analytical approach based on the Hidden Markovian Models (HMMs) to manage the defaults of the segmentation module. We also selected an optimal alphabet of graphemes in order to increase the performances of the recognition system. Furthermore, HMMs being developed exploit and model the notion of sub-words that is inherent to Arabic handwriting. An average correction of recognition rate of over 82.5% is obtained (in the first rank) with a lexicon of 232 different Tunisian state names.

A novel approach to extract data from check images is proposed based on the determination of baselines of checks, a priori information about the positions of data on checks, and a layout-driven item extraction method. Several techniques and algorithms have been developed in this approach including check image preprocessing, the extraction and identification of baselines, the extraction of the strokes of handwritten legal amounts, courtesy amounts and date, and the separation of strokes connected to baselines. A complete working system has been developed. The results of both testing experiments and on-line applications show that this approach is effective and the proposed techniques and algorithms perform well.

In most document analysis and recognition systems, straight lines are considered as one of the basic elements that should be located and eliminated to simplify the process of document analysis and recognition. The superposition or the intersection of different objects of interest found in the same area makes the process of detecting and extracting these line segments a non-trivial task to pursue, especially, if the method should preserve the valuable objects of interest that intersect with these lines. In this paper, we present a new and effective approach that detects the existence of line segments and eliminates them with the challenge of preserving the valuable information that intersects these line segments. The new approach makes use of the well-known morphological processing technique of the closing operation, that uses a fixed structuring element, towards the use of a dynamic structuring element. The purpose of the new dynamic structuring element is to detect and preserve the valuable objects intersecting the line segments that should be eliminated regardless of the different orientations with which the objects intersect these line segments.

We present a formal model for processing gray-scale images of business forms such as bank cheques. The formal model is based on a new hybrid-based approach namely the base lines. In fact, to segment handwritten and hand-printed data from bank cheques, knowledge rules and base lines will have important roles to segment and extract the information from bank cheques. The architectural design as well as the major components of the system is discussed in full detail. Moreover, the significant use of the morphological followed by the topological processing on gray-scale images is used as a major aspect to restore the lost information after the elimination of the background and the base lines from the gray-scale cheques.

This paper presents a new perception based model for reading cursive script. We describe the organization of our pseudo-neuronal system and show the role of activation mechanism in perceiving and reading cursive script. We have introduced into our model some characteristics specific to cursive script. First, we use more appropriate features such as ascenders and descenders. Second, we deal with the ambiguity of letter location by introducing the concept of the fuzzy position. The location as well as the missing letters are deduced from the context (i.e. the word-letter lexicon). After implementation of our method, preliminary qualitative results have been obtained and are discussed. We are concentrating now on further formalizing and generalizing the proposed model on a larger data base.

Millions of financial transactions take place every day. Associated with them are documents such as bank cheques, payment slips and bills which have to be processed. A great deal of time, effort and money will be saved if they can be entered into the computer and processed automatically. According to the specific characteristics of financial documents, it can be concluded that it is possible to build a system for recognizing specific types of financial documents, instead of a complex and general one aiming at different kinds of documents. In this paper, a financial document recognition prototype system which can process bank cheques, payment slips and bills, is presented. It consists of four major parts: (a) document image acquisition including scanning and binarization, (b) fixed document processing subsystem based on the detection of staff lines, (c) flexible document processing subsystem operating in a form description language (FDL), and (d) character recognition. Numerous experimental results are presented and discussed.>

We present a new method to segment visual handwritten data in gray-scale images. In handwriting recognition, visual shapes are very important in improving the system's performance. We introduce a robust method for extracting visual shapes of handwritten data from a noisy background. We adopted a multi-resolution Marr-Hildreth (1980) based approach to correctly segment visual data in variable contrasted images. Encouraging results have been obtained on real data, from the CEDAR database.>

Evaluates the extended shadow code used as a global feature vector for the signature verification problem. The latter as a shape factor is very easy to implement. Numerical experiments have been made with a signature database of 800 images (20 writers /spl times/ 40 signatures per writer) in the context of random forgeries. In the first experiment, a kNN classifier with voting shows a mean total error rate of 0.01% with k=1. In the second experiment, a minimum distance classifier has been used. The thresholds were evaluated for each of the 20 writers, and the number of reference signatures was varied in the range 1 /spl les/ N/sub ref/ /spl les/ 10. The mean total error rate was below 1.00% with N/sub ref/ = 4 genuine signatures.>

The authors propose a character segmentation algorithm which is a region-based approach using background pixels. Some interesting background regions are found automatically by performing two independent filtering steps: top-down filtering and bottom-up filtering. Relationships between the resulting background components are considered to select plausible regions favorable to segmentation. For preliminary experimentation, the authors collected 120 samples of connected digits written by 12 people, 97 pairs were successfully segmented.>