Federated learning (FL) is a collaborative learning scheme to train deep learning models, where collaborating parties can consolidate their models without sharing local data with other parties, hence preserving data privacy. Nevertheless, when implementing FL in Industrial visual inspection (IVI), the constraints posed by limited data availability and the intricate nature of the inspection tasks significantly impact the performance of the resulting model. This paper introduces FedTR, a novel FL framework incorporating transfer learning designed for Autonomous IVI, focusing on the challenging task of identifying label defects through end-to-end text recognition. Transfer learning is a method that leverages the knowledge of a pre-trained model to adapt to a different dataset. FedTR initially trains the model using a publicly available dataset, after which performs the essential federated learning process with model fine-tuning on the distributed and limited private data. Extensive experiment results demonstrate the effectiveness and feasibility of FedTR on private ink cartridge datasets for label defect identification. FedTR achieves an end-to-end text recognition word-level accuracy of 95.5% and 94.2% on homogeneous and heterogeneous data respectively. Additionally, it attains performance levels that are on par with those achieved through centralized training.

In this paper, we propose TECO, a multi-dimensional pruning framework to collaboratively prune the three dimensions (depth, width, and resolution) of convolutional neural networks (CNNs) for better execution efficiency on embedded hardware. In TECO, we first introduce a two-stage importance evaluation framework, which efficiently and comprehensively evaluates each pruning unit according to both the local importance inside each dimension and the global importance across different dimensions. Based on the evaluation framework, we present a heuristic pruning algorithm to progressively prune the three dimensions of CNNs towards the optimal trade-off between accuracy and efficiency. Experiments on multiple benchmarks validate the advantages of TECO over existing state-of-the-art (SOTA) approaches. The code and pre-trained models are available anonymously at https://github.com/ntuliuteam/Teco.

In this paper, we propose a multi-dimensional pruning framework, EMNAPE, to jointly prune the three dimensions (depth, width, and resolution) of convolutional neural networks (CNNs) for better execution efficiency on embedded hardware. In EMNAPE, we introduce a two-stage evaluation strategy to evaluate the importance of each pruning unit and identify the computational redundancy in the three dimensions. Based on the evaluation strategy, we further present a heuristic pruning algorithm to progressively prune redundant units from the three dimensions for better accuracy and efficiency. Experiments demonstrate the superiority of EMNAPE over existing methods.

Convolutional neural networks (CNNs) have demonstrated encouraging results in image classification tasks. However, the prohibitive computational cost of CNNs hinders the deployment of CNNs onto resource-constrained embedded devices. To address this issue, we propose EdgeCompress, a comprehensive compression framework to reduce the computational overhead of CNNs. In EdgeCompress, we first introduce dynamic image cropping (DIC), where we design a lightweight foreground predictor to accurately crop the most informative foreground object of input images for inference, which avoids redundant computation on background regions. Subsequently, we present compound shrinking (CS) to collaboratively compress the three dimensions (depth, width, and resolution) of CNNs according to their contribution to accuracy and model computation. DIC and CS together constitute a multidimensional CNN compression framework, which is able to comprehensively reduce the computational redundancy in both input images and neural network architectures, thereby improving the inference efficiency of CNNs. Further, we present a dynamic inference framework to efficiently process input images with different recognition difficulties, where we cascade multiple models with different complexities from our compression framework and dynamically adopt different models for different input images, which further compresses the computational redundancy and improves the inference efficiency of CNNs, facilitating the deployment of advanced CNNs onto embedded hardware. Experiments on ImageNet-1K demonstrate that EdgeCompress reduces the computation of ResNet-50 by 48.8% while improving the top-1 accuracy by 0.8%. Meanwhile, we improve the accuracy by 4.1% with similar computation compared to HRank. The state-of-the-art compression framework. The source code and models are available athttps://github.com/ntuliuteam/edge-compress.

Crossbar-based In-Memory Processing (IMP) accelerators have been widely adopted to achieve high-speed and low-power computing, especially for deep neural network (DNN) models with numerous weights and high computational complexity. However, the floating-point (FP) arithmetic is not compatible with crossbar architectures. Also, redundant weights of current DNN models occupy too many crossbars, limiting the efficiency of crossbar accelerators. Meanwhile, due to the inherent non-ideal behavior of crossbar devices, like write variations, pre-trained DNN models suffer from accuracy degradation when it is deployed on a crossbar-based IMP accelerator for inference. Although some approaches are proposed to address these issues, they often fail to consider the interaction among these issues, and introduce significant hardware overhead for solving each issue. To deploy complex models on IMP accelerators, we should compact the model and mitigate the influence of device non-ideal behaviors without introducing significant overhead from each technique. In this paper, we first propose to reuse bit-shift units in crossbars for approximately multiplying scaling factors in our quantization scheme to avoid using FP processors. Second, we propose to apply kernel-group pruning and crossbar pruning to eliminate the hardware units for data aligning. We also design a zerorize-recover training process for our pruning method to achieve higher accuracy. Third, we adopt the runtime-aware non-ideality adaptation with a self-compensation scheme to relieve the impact of non-ideality by exploiting the feature of crossbars. Finally, we integrate these three optimization procedures into one training process to form a comprehensive learning framework for co-optimization, which can achieve higher accuracy. The experimental results indicate that our comprehensive learning framework can obtain significant improvements over the original model when inferring on the crossbar-based IMP accelerator, with an average reduction of computing power and computing area by 100.02× and 17.37×, respectively. Furthermore, we can obtain totally integer-only, pruned, and reliable VGG-16 and ResNet-56 models for the Cifar-10 dataset on IMP accelerators, with accuracy drops of only 2.19% and 1.26%, respectively, without any hardware overhead.

For model privacy, local model parameters in federated learning shall be obfuscated before sent to the remote aggregator. This technique is referred to assecure aggregation. However, secure aggregation makes model poisoning attacks such as backdooring more convenient given that existing anomaly detection methods mostly require access to plaintext local models. This paper proposes a new federated learning technique SAFE-Learning to support backdoor detection for secure aggregation. We achieve this through two new primitives -oblivious random grouping (ORG)andpartial parameter disclosure (PPD). ORG partitions participants into one-time random subgroups with group configurations oblivious to participants; PPD allows secure partial disclosure of aggregated subgroup models for anomaly detection without leaking individual model privacy. ORG is based on our construction of several new primitives including tree-based random subgroup generation, oblivious secure aggregation, and randomized Diffie-Hellman key exchange. ORG can thwart colluding attackers from knowing each other’s group membership assignment with non-negligible advantage than random guess. Backdoor attacks are detected based on statistical distributions of the subgroup aggregated parameters of the learning iterations. SAFELearning can significantly reduce backdoor model accuracy without jeopardizing the main task accuracy under common backdoor strategies. Extensive experiments show SAFELearning is robust against malicious and faulty participants, whilst being more efficient than the state-of-art secure aggregation protocol in terms of both communication and computation costs.

Scaling down the resolution of input images can greatly reduce the computational overhead of convolutional neural networks (CNNs), which is promising for edge AI. However, as an image usually contains much spatial redundancy, e.g., background pixels, directly shrinking the whole image will lose important features of the foreground object and lead to severe accuracy degradation. In this paper, we propose a dynamic image cropping framework to reduce the spatial redundancy by accurately cropping the foreground object from images. To achieve the instance-aware fine cropping, we introduce a lightweight foreground predictor to efficiently localize and crop the foreground of an image. The finely cropped images can be correctly recognized even at a small resolution. Meanwhile, computational redundancy also exists in CNN architectures. To pursue higher execution efficiency on resource-constrained embedded devices, we also propose a compound shrinking strategy to coordinately compress the three dimensions (depth, width, resolution) of CNNs. Eventually, we seamlessly combine the proposed dynamic image cropping and compound shrinking into a unified compression framework, Smart Scissor, which is expected to significantly reduce the computational overhead of CNNs while still maintaining high accuracy. Experiments on ImageNet-1K demonstrate that our method reduces the computational cost of ResNet50 by 41.5% while improving the top-1 accuracy by 0.3%. Moreover, compared to HRank, the state-of-the-art CNN compression framework, our method achieves 4.1% higher top-1 accuracy at the same computational cost. The codes and data are available at https://github.com/ntuliuteam/smart-scissor

Federated Learning (FL) empowers multiple clients to collaboratively learn a model, enlarging the training data of each client for high accuracy while protecting data privacy. However, when deploying FL in real-time edge systems, the heterogeneity of devices among systems has a severe impact on the performance of the inferred model. Existing optimizations on FL focus on improving the training efficiency but fail to speed up inference, especially when there is a latency constraint. In this work, we propose Collate, a novel training framework that collaboratively learns heterogeneous models to meet the latency constraints of multiple edge systems simultaneously. We design a dynamic zeroizing-recovering method to adjust each local model architecture for high accuracy under its latency constraint. A proto-corrected federated aggregation scheme is also introduced to aggregate all heterogeneous local models, satisfying the latency constraint of different systems with only one training process and maintaining high accuracy. Extensive experiments indicate that, compared to state-of-the-art methods and under a latency constraint, our extended models can improve the accuracy by 1.96% on average, and our shrunk models can also obtain a 3.09% accuracy improvement on average, with almost no extra training overhead. The related codes and data will be available at https://github.com/ntuliuteam/Collate.

This paper proposes an exact model for the planning and design problem of fog networks. More precisely, a mathematical model is proposed to simultaneously determine the optimal location, the capacity and the number of fog node(s) as well as the interconnection between the installed fog nodes and the cloud. The goal of the model is to minimize the delay in the network and the amount of traffic sent to the cloud data center. To address this multi-objective optimization problem, three optimization techniques are used: the weighted sum, the hierarchical and the trade-off methods. The weighted sum method aggregates all the lone objective functions into a single objective by applying a weighted vector. The hierarchical method takes a sequential approach by tightly constraining the more important objective function. The trade-off method solves a single objective function and translates all other objective functions into constraints. These methods are then compared in terms of average delay, amount of traffic sent to the cloud and amount of CPU time required to find optimal solution(s). Since we are dealing with a multi-objective optimization problem and that multiple optimal solutions can be found, the fuzzy-based mechanism and the hypervolume indicator have been used. Computational results show that as the problem size increases, the delay and the traffic also increase in a linear form; whereas, the solution time increases in non-polynomial time. The weighted sum method was able to achieve the best trade-off results for the delay and the traffic, whereas the hierarchical method was able to return minimum delay but with worse traffic going to the cloud. As the model considers realistic edge device traffic parameters, constraints, and various topology aspects, it can be helpful for the planning and deployment of fog networks and how they operate within a cloud infrastructure.

Policy-based access control systems are crucial for secure information sharing in collaborative applications. However, policy management needs to be flexible in order to adapt to different environments and be able to support policy evolution. However, when dealing with large sets of evolving policies, it is critical that policies meet certainpolicy quality requirements. Policy sets must be complete, free of inconsistencies, and relevant. In this paper, we propose a framework to analyze policies to determine whether they meet such requirements. Our framework uses provenance techniques to collect comprehensive data about actions which were either triggered due to a network context or a user (i.e., a human or a device) action. The framework includes two approaches for policy analysis: structure-based and classification-based. For the structure-based approach, we designed tree structures to organize and assess the policy set efficiently. For the classification-based approach, we employed the classification techniques to learn the characteristics of policies and predict their quality. In addition, the framework includes the policy evolution module which mainly consists of recommendation and re-evaluation services for policy changes which both aim at fulfilling the policy quality requirements. The analysis framework has been implemented and experimental results from the prototype are reported.

Fog computing has risen as a promising technology for augmenting the computational and storage capability of the end devices and edge networks. The urging issues in this networking paradigm are fog nodes planning, resources allocation, and offloading strategies. This paper aims to formulate a mathematical model which jointly tackles these issues. The goal of the model is to optimize the tradeoff (Pareto front) between the capital expenditure and the network delay. To solve this multiobjective optimization problem and obtain benchmark values, we first use the weighted sum method and two existing evolutionary algorithms (EAs), nondominated sorting genetic algorithm II and speed-constrained multiobjective particle swarm optimization. Then, inspired by those EAs, this paper proposes a new EAs, named particle swarm optimized nondominated sorting genetic algorithm, which combines the convergence and searching efficiency of the existing EAs. The effectiveness of the proposed algorithm is evaluated by the hypervolume and inverted generational distance indicators. The performance evaluation results show that the proposed model and algorithms can help the network planners in the deployment of fog networks to complement their existing computation and storage infrastructure.

Emerging technologies and applications including Internet of Things, social networking, and crowd-sourcing generate large amounts of data at the network edge. Machine learning models are often built from the collected data, to enable the detection, classification, and prediction of future events. Due to bandwidth, storage, and privacy concerns, it is often impractical to send all the data to a centralized location. In this paper, we consider the problem of learning model parameters from data distributed across multiple edge nodes, without sending raw data to a centralized place. Our focus is on a generic class of machine learning models that are trained using gradient-descent-based approaches. We analyze the convergence bound of distributed gradient descent from a theoretical point of view, based on which we propose a control algorithm that determines the best tradeoff between local update and global parameter aggregation to minimize the loss function under a given resource budget. The performance of the proposed algorithm is evaluated via extensive experiments with real datasets, both on a networked prototype system and in a larger-scale simulated environment. The experimentation results show that our proposed approach performs near to the optimum with various machine learning models and different data distributions.

Emerging technologies and applications including Internet of Things (IoT), social networking, and crowd-sourcing generate large amounts of data at the network edge. Machine learning models are often built from the collected data, to enable the detection, classification, and prediction of future events. Due to bandwidth, storage, and privacy concerns, it is often impractical to send all the data to a centralized location. In this paper, we consider the problem of learning model parameters from data distributed across multiple edge nodes, without sending raw data to a centralized place. Our focus is on a generic class of machine learning models that are trained using gradient-descent based approaches. We analyze the convergence rate of distributed gradient descent from a theoretical point of view, based on which we propose a control algorithm that determines the best trade-off between local update and global parameter aggregation to minimize the loss function under a given resource budget. The performance of the proposed algorithm is evaluated via extensive experiments with real datasets, both on a networked prototype system and in a larger-scale simulated environment. The experimentation results show that our proposed approach performs near to the optimum with various machine learning models and different data distributions.

Successful collaborations require information and resource sharing and thus adequate access control policy management systems that control sharing among the collaborating entities. Such management systems need to be flexible in order to adapt to different environments and thus be able to support access control policy evolution. However, when dealing with large sets of evolving policies it is critical that policies meet certain "policy quality requirements". Specifically, policies of interest must be up-to-date, complete, free of inconsistencies, relevant. In this paper, we propose an approach to analyze policies in order to determine whether policies meet such requirements. Our approach is based on the use of provenance techniques that collect comprehensive data about actions executed by users in the context of workflows, that is, sets of tasks executed according to some ordering by users. Provenance data are used by services that support various types of analysis to determine whether the policies of interest verify the quality requirements.

Proxy Mobile IPv6 (PMIPv6) has been developed by the IETF as a network-based mobility management protocol to support the mobility of IP devices. F-PMIPv6 and Localized Routing have been proposed to solve the issues of packet loss/handover delay and non-optimal routing of data packets respectively. Optimized PMIPv6 (O-PMIPv6), as proposed in [1], looked at combining all these features in one solution and was initially designed to work in predictive mode. However, when the mobile node moves very fast, the attachment with the target network may happen before the completion of the tunnel setup procedure between the previous and the new access network while using predictive mode. Moreover, the prediction of the attachment could fail for several reasons such as the lost of the detachment notification messages. All of these scenarios require support of reactive mode for O-PMIPv6. As a result, this paper extends the operation of O-PMIPv6 to the reactive mode. A comparative analysis is done with PMIPv6 and the reactive mode of F-PMIPv6 to study the effectiveness of the proposed extension in terms of route optimization handover delay, signaling cost and network utilization. Finally, the reactive O-PMIPv6 is compared to the predictive O-PMIPv6 to show differences in performance for both approaches.

Proxy Mobile IPv6 (PMIPv6) has been developed by the IETF as a network-based mobility management protocol to support the mobility of IP devices. Although several proposals have been made for localized routing optimization, they don't take into account handover management and localized routing simultaneously. In fact, the localized routing state is only restored after the handover, leading to packet loss and signaling overhead. On the other hand, Fast Handovers for PMIPv6 (F-PMIPv6) protocol has been designed to mainly solve the issues of long handover delay and packets loss during handover. As a result, this paper looks at enhancing F-PMIPv6 by combining the handover with route optimization by proposing a new protocol called Optimized Proxy Mobile IPv6 (O-PMIPv6). The proposed protocol enhances the performance of PMIPv6 and F-PMIPv6 in terms of route optimization handover delay, signaling cost, and network utilization.

The NGN (Next Generation Network), which can provide advanced multimedia services over an all-IP based network, has been the subject of considerable attention for several years. While there have been tremendous efforts to develop its architecture and protocols, especially for IMS, a key technology of the NGN, its wide deployment is still a long way off. However, efforts to create an advanced signaling infrastructure able to meet many requirements have resulted in a large number of functional components and interactions between these components. Thus, the carriers are trying to explore effective ways to deploy IMS while offering value-added services. As one such approach, we have proposed a self-organizing IMS. A self-organizing IMS enables IMS functional components and corresponding physical nodes to adapt dynamically and automatically based on actual conditions such as network load and available system resources while continuing IMS operation. To realize this, service continuity for users is an important requirement when a reconfiguration occurs during operation. In this paper, we propose a mechanism that will provide service continuity to users without any impact on clients by extending the SBC (Session Border Controller). Furthermore, we implement it, show its behavior and evaluate its processing time.

Mobility management with provision of seamless handover is crucial for an efficient support of global roaming of mobile nodes (MNs) in next-generation wireless networks (NGWN). Mobile IPv6 (MIPv6) and its extensions were proposed by IETF for IP layer mobility management. However, performance of IPv6-based mobility management schemes is highly dependent on traffic characteristics and user mobility models. Consequently, it is important to assess this performance in-depth through those two factors. The performance of IPv6-based mobility management schemes is usually evaluated through simulations. This paper proposes an analytical framework to evaluate the performance of IPv6-based mobility management protocols. This proposal does not aim to advocate which is better but rather to study the effects of various network parameters on the performance of these protocols to enlighten decision-making. The effect of system parameters, such as subnet residence time, packet arrival rate and wireless link delay, is investigated for performance evaluation with respect to various metrics like signaling overhead cost, handoff latency and packet loss. Numerical results show that there is a trade-off between performance metrics and network parameters.

Mobility management, with provision of seamless handoff and quality of service (QoS) guarantees, is one of the key issues in next generation or 4G wireless networks (NGWN/4G). Current trends in communication networks evolution are directed towards an all-IP principles in order to hide heterogeneities and to achieve convergence of various access networks. Several IPv6-based mobility management schemes have been proposed for service continuity in NGWN/4G. However, these schemes have some well-known disadvantages such as signaling traffic overhead, high packet loss and high handoff latency, thereby causing a user-perceptible deterioration of real-time applications. This paper proposes an efficient handoff protocol for NGWN/4G, called handoff protocol for integrated networks (HPIN), that alleviates service disruption during handoff. HPIN is a one suite protocol that performs local mobility management, fast handoff, context transfer and access network discovery. Performance evaluation based on numerical results shows that the proposed scheme performs better than existing schemes.