Currently, edge Artificial Intelligence (AI) systems have significantly facilitated the functionalities of intelligent devices such as smartphones and smart cars, and supported diverse applications and services. This fundamental supports come from continuous data analysis and computation over these devices. Considering the resource constraints of terminal devices, multi-layer edge artificial intelligence systems improve the overall computing power of the system by scheduling computing tasks to edge and cloud servers for execution. Previous efforts tend to ignore the nature of strong pipelined characteristics of processing tasks in edge AI systems, such as the encryption, decryption and consensus algorithm supporting the implementation of Blockchain techniques. Therefore, this paper proposes a new pipelined task scheduling algorithm (referred to as PTS-RDQN), which utilizes the system representation ability of deep reinforcement learning and integrates multiple dimensional information to achieve global task scheduling. Specifically, a co-optimization strategy based on Rainbow Deep Q-Learning (RainbowDQN) is proposed to allocate computation tasks for mobile devices, edge and cloud servers, which is able to comprehensively consider the balance of task turnaround time, link quality, and other factors, thus effectively improving system performance and user experience. In addition, a task scheduling strategy based on PTS-RDQN is proposed, which is capable of realizing dynamic task allocation according to device load. The results based on many simulation experiments show that the proposed method can effectively improve the resource utilization, and provide an effective task scheduling strategy for the edge computing system with cloud-edge-end architecture.

The inefficient utilization of ubiquitous graph data with combinatorial structures necessitates graph embedding methods, aiming at learning a continuous vector space for the graph which is amenable to be adopted in traditional machine learning algorithms in favor of vector representations. Graph embedding methods build an important bridge between social network analysis and data analytics as social networks naturally generate an unprecedented volume of graph data continuously. Publishing social network data not only bring benefit for public health, disaster response, commercial promotion, and many other applications, but also give birth to threats that jeopardize each individual’s privacy and security. Unfortunately, most existing works in publishing social graph embedding data only focus on preserving social graph structure with less attention paid to the privacy issues inherited from social networks. To be specific, attackers can infer the presence of a sensitive relationship between two individuals by training a predictive model with the exposed social network embedding. In this paper, we propose a novel link-privacy preserved graph embedding framework using adversarial learning, which can reduce adversary’s prediction accuracy on sensitive links while persevering sufficient non-sensitive information such as graph topology and node attributes in graph embedding. Extensive experiments are conducted to evaluate the proposed framework using ground truth social network datasets.

Graph data publication has been considered as an important step for data analysis and mining. Graph data, which provide knowledge on interactions among entities, can be locally generated and held by distributed data owners. These data are usually sensitive and private, because they may be related to owners’ personal activities and can be hijacked by adversaries to conduct inference attacks. Current solutions either consider private graph data as centralized contents or disregard the overlapping of graphs in distributed manners. Therefore, this work proposes a novel framework for distributed graph publication. In this framework, differential privacy is applied to justify the safety of the published contents. It includes four phases, i.e., graph combination, plan construction sharing, data perturbation, and graph reconstruction. The published graph selection is guided by one data coordinator, and each graph is perturbed carefully with the Laplace mechanism. The problem of graph selection is formulated and proven to be NP-complete. Then, a heuristic algorithm is proposed for selection. The correctness of the combined graph and the differential privacy on all edges are analyzed. This study also discusses a scenario without a data coordinator and proposes some insights into graph publication.