Federated learning (FL) enables geographically dispersed edge devices (i.e., clients) to learn a global model without sharing the local datasets, where each client performs gradient descent with its local data and uploads the gradients to a central server to update the global model. However, FL faces massive communication overhead resulted from uploading the gradients in each training round. To address this problem, most existing research compresses the gradients with fixed and unified quantization for all the clients, which neither seeks adaptive quantization due to the varying gradient norms at different rounds, nor exploits the heterogeneity of the clients to accelerate FL. In this paper, we propose a novel adaptive and heterogeneous gradient quantization algorithm (AdaGQ) for FL to minimize the wall-clock training time from two aspects: i) adaptive quantization which exploits the change of gradient norm to adjust the quantization resolution in each training round; and ii) heterogeneous quantization which assigns lower quantization resolution to slow clients to align their training time with other clients to mitigate the communication bottleneck, and higher quantization resolution to fast clients to achieve a better communication efficiency and accuracy tradeoff. Evaluations based on various models and datasets validate the benefits of AdaGQ, reducing the total training time by up to 52.1% compared to baseline algorithms (e.g., FedAvg, QSGD).

Graphics processing units (GPUs) are the de facto standard for processing deep learning (DL) tasks. In large-scale GPU clusters, GPU failures are inevitable and may cause severe consequences. For example, GPU failures disrupt distributed training, crash inference services, and result in service level agreement violations. In this paper, we study the problem of predicting GPU failures using machine learning (ML) models to mitigate their damages.

As IoT devices are becoming widely deployed, there exist many threats to IoT-based systems due to their inherent vulnerabilities. One effective approach to improving IoT security is to deploy IoT honeypot systems, which can collect attack information and reveal the methods and strategies used by attackers. However, building high-interaction IoT honeypots is challenging due to the heterogeneity of IoT devices. Vulnerabilities in IoT devices typically depend on specific device types or firmware versions, which encourages attackers to perform pre-attack checks to gather device information before launching attacks. Moreover, conventional honeypots are easily detected because their replying logic differs from that of the IoT devices they try to mimic.To address these problems, we develop an adaptive high-interaction honeypot for IoT devices, called em HoneyIoT. We first build a real device based attack trace collection system to learn how attackers interact with IoT devices. We then model the attack behavior through markov decision process and leverage reinforcement learning techniques to learn the best responses to engage attackers based on the attack trace. We also use differential analysis techniques to mutate response values in some fields to generate high-fidelity responses.HoneyIoT has been deployed on the public Internet. Experimental results show that HoneyIoT can effectively bypass the pre-attack checks and mislead the attackers into uploading malware. Furthermore, HoneyIoT is covert against widely used reconnaissance and honeypot detection tools.

Video analytics demand intensive computation resources, which means long processing delay when running on mobile devices. Although offloading computation to the cloud can partially solve the problem, transferring videos to the cloud introduces high transmission delay. With mobile edge computing, computation can be offloaded to the nearby edge servers to reduce the delay. However, the computation resources of the edge servers are usually limited and highly dynamic, and then server selection should be adaptive in order to improve the performance of video analytics. Also, frame resolution should be selected to achieve a better tradeoff between accuracy and frame processing rate. In this paper, we study the server resource-aware offloading problem for video analytics, where the goal is to maximize the utility which is a weighted function of accuracy and frame processing rate. The major challenge to solve this problem is the lack of server and network knowledge and the dynamic system environment. To overcome these challenges, we formulate the problem as a contextual Multi-armed Bandit problem, and propose a Bayesian Optimization based online learning algorithm to gradually learn the server status and the optimal solution, and make it adaptable for time-varying environments. Both theoretical analysis and evaluation results demonstrate the superior performance of our proposed algorithm.

Visible light communication (VLC) networks can provide high‐rate transmissions with advantages such as high security, low power consumption and so on. Multiple users get access to the VLC network according to the carrier sense multiple access with collision avoidance (CSMA/CA) mechanism defined by IEEE 802.15.7 media access control layer specifications. In this study, the authors propose an energy efficient medium access scheme to improve the performance of IEEE 802.15.7 CSMA/CA mechanism under unsaturated traffic conditions. Aiming at reducing the power consumed on unnecessary waiting and at improving the channel utilisation efficiency, they first apply a successive transmission scheme on the node, which has completed one successful transmission and gets the chance of transmitting the next packet without backoff delay period according to the network condition. Then, a dynamic contention window is exploited to mitigate the collisions, where they derive the optimal minimum contention window based on throughput analysis. The power consumption can be reduced with improved throughput, when both the traffic load and the number of nodes in the network are taken into account. Simulation results demonstrate that the authors’ proposed scheme outperforms the IEEE 802.15.7 CSMA/CA scheme with respect to the average power consumption, system throughput and packet dropping probability.