In recent years, convolutional neural networks have achieved state-of-the-art performance in a number of computer vision problems such as image classification. Prior research has shown that a transfer learning technique known as parameter fine-tuning wherein a network is pre-trained on a different dataset can boost the performance of these networks. However, the topic of identifying the best source dataset and learning strategy for a given target domain is largely unexplored. Thus, this research presents and evaluates various transfer learning methods for fine-grained image classification as well as the effect on ensemble networks. The results clearly demonstrate the effectiveness of parameter fine-tuning over random initialization. We find that training should not be reduced after transferring weights, larger, more similar networks tend to be the best source task, and parameter fine-tuning can often outperform randomly initialized ensembles. The experimental framework and findings will help to train models with improved accuracy.

Reputation systems provide an important basis for judging whether to interact with others in a networked system. Designing such systems is especially interesting in decentralized environments such as mobile ad-hoc networks, as there is no trusted authority to manage reputation feedback information. Such systems also come with significant privacy concerns, further complicating the issue. Researchers have proposed privacy-preserving decentralized reputation systems (PDRS) which ensure individual reputation information is not leaked. Instead, aggregate information is exposed. Unfortunately, in existing PDRS, when a party leaves the network, all of the reputation information they possess about other parties in the network leaves too. This is a significant problem when applying such systems to the kind of dynamic networks we see in mobile computing. In this article, we introduce dynamic, privacy-preserving reputation systems (Dyn-PDRS) to solve the problem. We enumerate the features that a reputation system must support in order to be considered a Dyn-PDRS. Furthermore, we present protocols to enable these features and describe how our protocols are composed to form a Dyn-PDRS. We present simulations of our ideas to understand how a Dyn-PDRS impacts information availability in the network, and report on an implementation of our protocols, including timing experiments.

The hypothesis of this paper is that topics, expressed through large-scale social media networks, approximate electricity utilization events (e.g., using high power consumption devices such as a dryer) with high accuracy. Traditionally, researchers have proposed the use of smart meters to model device-specific electricity utilization patterns. However, these techniques suffer from scalability and cost challenges. To mitigate these challenges, we propose a social media network-driven model that utilizes large-scale textual and geospatial data to approximate electricity utilization patterns, without the need for physical hardware systems (e.g., such as smart meters), hereby providing a readily scalable source of data. The methodology is validated by considering the problem of electricity use disaggregation, where energy consumption rates from a nine-month period in San Diego, coupled with 1.8 million tweets from the same location and time span, are utilized to automatically determine activities that require large or small amounts of electricity to accomplish. The system determines 200 topics on which to detect electricity-related events and finds 38 of these to be valid descriptors of energy utilization. In addition, a comparison with electricity consumption patterns published by domain experts in the energy sector shows that our methodology both reproduces the topics reported by experts, while discovering additional topics. Finally, the generalizability of our model is compared with a weather-based model, provided by the U.S. Department of Energy.

The cognitive radio (CR) field currently lacks a standardized end-user test methodology that is repeatable, flexible, and effective across multiple CR architectures. Furthermore, the CR field lacks a suitable device-agnostic framework that allows testing of an integrated CR system and not solely specific components. This paper presents a CR test methodology, known as the Cognitive RAdio Test Methodology (CRATM), to address these issues. CRATM proposes to use behavior-based testing in which cognition may be measured by evaluating both primary user (PU) and secondary user (SU) (i.e., the CR under test) performance. Data on behavior-based testing are collected and evaluated. An SU pair and a PU radio pair are implemented using the Wireless Open-Access Research Platform (WARP) and the WARPLab software running in MATLAB. The PU is used to create five distinct radio-frequency environments utilizing narrow-band, wideband, and noncontiguous waveforms. The SU response to the PU created environments is measured. The SU implements a simple cognitive engine (CE) that incorporates energy-detection spectrum sensing. The effect of the CE on both SU and PU performance is measured and evaluated.

With the success and ubiquity of large scale, social media networks comes the challenge of assessing the veracity of information shared across them that inform individuals about emerging real-world events and trends. We propose a veracity-assessment model for information dissemination on social media networks that combines natural language processing and machine learning algorithms to mine textual content generated by each user. Large scale social media networks (such as Twitter and Facebook) are considered digital communication platforms, in which information can be quickly and easily exchanged, thereby expanding the breadth of knowledge across the globe. In this paper, four case studies spanning multiple geographic regions, threat scenarios and time frames are investigated, in order to demonstrate how real-world events impact the manner in which information/misinformation is communicated and spread through a social media network. Our results show that metadata associated with each user can provide significant insight on the social media network's users' tendency to accurately discuss a topic.

The radio frequency (RF) spectrum is a limited resource. Spectrum allotment disputes stem from this scarcity as many radio devices are confined to a fixed frequency. One alternative is to incorporate reconfigurability within a cognitive radio platform, thereby enabling the radio to adapt to dynamic RF spectrum environments. In this way, the radio is able to actively observe the RF spectrum, orient itself to the current RF environment, decide on a mode of operation, and act accordingly, thereby sharing the spectrum and operating in more a flexible manner. This research presents a novel architecture for the purpose of adapting radio operation to the current RF spectrum environment. Specifically, this research makes three contributions: (1) a framework for testing and evaluating clustering algorithms in the context of cognitive radio networks, (2) a new RF spectrum map merging technique for adaptive waveform selection, with initial integration testing on a field-programmable gate array (FPGA), and (3) a novel cognitive radio network emulation framework for testing and evaluating totally-ordered multicast as a means for inter-node communication.

The smart grid offers an exciting way to better manage various aspects of a given utility. This comes with an increased threat of privacy due to fine-grained information reporting from each smart meter. Privacy preserving protocols have been proposed in the literature to deal with this problem. In this paper, we look at optimizing one technique, secure multiparty computation (MPC), for application to in-network, privacy preserving computation for smart meter networks. We propose a new construction for secure MPC, which we call transferable MPC. We present protocols for this construction, and show how this leads to more efficient and scalable multiparty computations in a smart metering application.

The ever-increasing size and complexity of cloud computing, data centers, virtualization, web services, and other forms of distributed computing make automated and effective service management increasingly important. This article treats the service placement problem as a novel generalization of the on-line vector packing problem. This generalization of the service placement problem does not require a priori knowledge of the service resource profiles, allows for resource profiles to change over time, and allows services to be moved once placed on a server. An on-line self-organizing model profiles resource supplies and demands arranging services in a placement based on their resulting quality rating. A policy-driven asymmetric matrix norm quantifies the quality of the placement allowing for administrative preferences regarding service performance versus service inclusion. Service resource usage profiles' variations cause changes in their assigned placement quality; forcing new, better server placements to be found. Because some placements perform better, a proportional integral derivative controller for performance feedback adjusts the services' actual profile according to service's individual response times. This large scale system autonomically organizes placement of services in response to changes in demand and network disruptions. This article presents theorems which demonstrate the theoretical basis for the model. The article includes empirical results from the implementation of this model in a self-organizing testbed of web servers and services.

Although progress has been made to satisfy mobility at the network's edge, much work remains for deeper mobility requirements. In many emergency or military applications, fixed network infrastructure may not be available or even possible. As mobile military data requirements grow, and spectrum limitations increase, systems must better utilise the diminishing available bandwidth of wireless radio frequency and free space optical links. To achieve better utilisation, disruption-prone networks require disruption-tolerant protocols or localised buffering to mask disruptions. Transmission control protocol (TCP)/IP assumes reliable links and performs well in networks with congestion dominated packet losses, but poorly with link failure dominated packet losses. Although TCP might be altered for disruptive environments, evolutionary reasons make it difficult to do so well without partitioning networks into reliable and disruption-tolerant systems. Instead, the authors examine transport layer aware helper protocols, with intermediate buffering in routers, to assist TCP across disruption-prone network portions. The buffering does not require TCP modifications at communicating nodes and integrate well with existing routers (i.e. TCP friendly). Experimental results show that TCP can reliably establish and maintain connections under poor link availability using the buffering protocol. Few TCP connections complete without it.

Evolutionary algorithms can be applied to a variety of constrained network communication problems with centric type models. This paper shows that with real-world complex network communication problems of this type, sophisticated statistical search is required. This situation occurs due to the fact that these optimization problems are at least NP-complete. In order to appreciate the formal modeling of realistic communication networks, historical network design problems are presented and evolved into more complex real-world models with associated deterministic and stochastic solution approaches discussed. This discussion leads into the design of an innovative multiobjective evolutionary algorithm (MOEA) to solve a very complex network design problem variation called the multicommodity capacitated network design problem (MCNDP). This variation represents a hybrid real-world communication architecture as reflected in real-world network centric models with directional communications, multiple objectives including costs, delays, robustness, vulnerability, and operating reliability within network constraints. Nodes in such centric systems can have multiple and varying link capacities as well as rates and information (commodity) quantities to be sent and received. Each commodity can also have independent prioritized bandwidth and spectrum requirements. The nondominated sorting genetic algorithm (NSGA-II) is selected as the MOEA framework which is modified and parallelized to solve the generic MCNDP. Since the MCNDP is highly constrained but with an enormous number of possible network communication topologies, a novel initialization procedure and mutation method are integrated resulting in reduced search space. Empirical results and analysis indicate that effective topological Pareto solutions are generated for use in highly constrained, communication-based network design.

A mobile ad hoc network (MANET) consists of a group of communicating hosts that form an arbitrary network topology by means of any of several wireless communication media. MANET communications represent a diversification in communication technology necessary to solve the stringent end-to-end requirements of QoS-based communication networks. Of the many challenges in this complex distributed system, the problem of routing based on a predefined set of customer preferences, critical to guaranteeing quality-of-service, is the focus of this research. Specifically, this paper modifies a cluster-based QoS routing algorithm for mobile ad hoc networks with the aim of providing fault tolerance, which is a critical feature in providing QoS in the link failure-prone environment of mobile networks. Performance of this new fault-tolerant cluster-based QoS wireless algorithm is evaluated according to failure recovery time, dropped packets, throughput, and sustained flow bandwidth via simulations involving node failure scenarios along QoS paths.

This work defines a heuristic to determine the most optimal configuration for a service oriented infrastructure by effectively and efficiently matching network nodes' resource provisions to services' resource demands. The heuristic is a numerical method that attempts to find the global minima of a configuration quality metric. The efficiency and effectiveness of the heuristic is compared to simulated annealing in a large scale simulation. In the large scale, the heuristic is an order of magnitude better in both effectiveness and efficiency.

Gossip-based communication protocols are attractive in cases where absolute delivery guarantees are not required due to their scalability, low overhead, and probabilistically high reliability. In earlier work, a gossip-based protocol known as gravitational gossip was created that allows the selection of quality ratings within subgroups based on workload and information update frequency. This paper presents an improved protocol that adds an adaptive component that matches the actual subgroup communication rates with desired rates coping with network variations by modifying underlying gossip weights. The protocol is designed for use in environments where many information streams are being generated and interest levels vary between nodes in the system. The gossip-based protocol is able to allow subscribers to reduce their expected workload in return for a reduced information rate. The protocol is a good fit for applications such as military information systems, sensor networks, and rescue operations. Experiments were conducted in order to compare the merits of different adaptation mechanisms. Experimental results show promise for this approach.

Evolutionary algorithms have been applied to a variety of network flow problems with acceptable results. In this research, a multiobjective evolutionary algorithm (MOEA) is used to solve a variation of the multicommodity capacitated network design problem (MCNDP). This variation represents a hybrid communication network as found in network centric models with multiple objectives including costs, delays, robustness, vulnerability, and reliability. Nodes in such centric systems can have multiple and varying link capacities, rates and information (commodity) quantities to be delivered and received. Each commodity can have an independent prioritized bandwidth requirement as well. Insight to the MCNDP problem domain and Pareto structure is developed. The nondominated sorting genetic algorithm (NSGA-II) is modified and extended to solve such a MCNDP. Since the MCNDP is highly constrained, a novel initialization procedure and mutation method are also integrated into this MOEA. Empirical results and analysis indicate that effective solutions are generated very efficiently

No abstract available.