Edge computing is a key enabling technology that meets continuously increasing requirements for the intelligent Internet of Things (IoT) applications. To cope with the increasing privacy leakages of machine learning while benefiting from unbalanced data distributions, federated learning has been wildly adopted as a novel intelligent edge computing framework with localized training mechanism. However, recent studies found that the federated learning framework exhibits inherent vulnerabilities on active attacks, and poisoning attack is one of the most powerful and secluded attacks where the functionalities of the global model could be damaged through attackers well-crafted local updates. In this paper, we give a comprehensive exploration of the poisoning attack mechanisms in the context of federated learning. We first present a poison data generation method, named Data_Gen, based on the generative adversarial networks (GAN). This method mainly relies upon the iteratively updated global model parameters to regenerate samples of interested victims. Secondly, we further propose a novel generative poisoning attack model, named PoisonGAN, against the federated learning framework. This model utilizes the designed Data_Gen method to efficiently reduce the attack assumptions, and make attacks feasible in practice. We finally evaluate our data generation and attack models by implementing two-types of typical poisoning attack strategies, label-flipping and backdoor, on a federated learning prototype. The experimental results demonstrate that these two attack models are effective in federated learning.

In recent years, constant developments in Internet of Things (IoT) generate large amounts of data, which put pressure on Cloud computing’s infrastructure. The proposed Fog computing architecture is considered the next generation of Cloud Computing for meeting the requirements posed by the device network of IoT. One of the obstacles of Fog Computing is distribution of computing resources to minimize completion time and operating cost. The following study introduces a new approach to optimize task scheduling problem for Bag-of-Tasks applications in Cloud–Fog environment in terms of execution time and operating costs. The proposed algorithm named TCaS was tested on 11 datasets varying in size. The experimental results show an improvement of 15.11% compared to the Bee Life Algorithm (BLA) and 11.04% compared to Modified Particle Swarm Optimization (MPSO), while achieving balance between completing time and operating cost.

With the appearance of more and more devices connected to the Internet, the world has witnessed an ever-growing number of data to be processed. Among those, many tasks require swift execution time, while the storage and computation capability of Internet of Things (IoT) devices are limited. To address the demands of delay-sensitive tasks, we present a vehicular edge–cloud computing (VECC) network that leverages powerful computation capabilities through the deployment of servers in proximity to task-generated devices, as well as the utilization of idle resources from smart vehicles to share the workload. Because these limited resources are vulnerable to sudden data arising, it is imperative to incorporate cloud servers to prevent system overload. The challenge now is to find a task offloading strategy that collaborates both edges and cloud resources to minimize the total time surpassing the quality baseline of each task (tolerance time) and make all tasks meet their soft deadlines of quality. To reach this goal, we first model the task offloading problem in VECC as a Markov decision process (MDP). Then, we propose advantage-oriented task offloading with a dueling actor-insulator network scheme to solve the problem. This value-based reinforcement learning (RL) method helps the agent find an effective policy when not knowing all the state attributes changes. The effectiveness of our method is demonstrated by performance evaluations based on real-world bus traces in Rio de Janeiro (Brazil). The experimental results show that our proposal reduces the tolerance time by at least 8.81% compared to other RL algorithms and 75% compared to greedy approaches.