6G technology enables AI-based massive IoT to manage network resources and data with ultra high speed, responsive network and wide coverage. However, many artificial intelligence-enabled internet of things (AIoT) systems are vulnerable to adversarial example attacks. Therefore, designing robust deep learning models that can be deployed on resource-constrained devices has become an important research topic in the field of 6G-enabled AIoT. In this paper, we propose a method for automatically searching for robust and efficient neural network structures for AIoT systems. By introducing a skip connection structure, a feature map with reduced front-end influence can be used for calculations during the classification process. Additionally, a novel type of of dense connected search space is proposed. By relaxing this space, it is possible to search for network structures efficiently. In addition, combined with adversarial training and model delay constraints, we propose a multi-objective gradient optimization method to realize the automatic searching of network structures. Experimental results demonstrate that our method is effective for AIoT systems and superior to state-of-the-art neural architecture search algorithms.

Collaborative learning with multiple edge devices to build group intelligence is a new trend. Edge artificial intelligence (AI) computing often makes full use of various available data and resources in terminal devices, edge servers, and cloud data centers to achieve collaborative deci-sion-making. However, in order to achieve the goal, it should also guarantee the security of data storage, transmission, and management. Therefore, how to build a distributed trust system for cooperative learning in edge computing is a complex and challenging engineering problem. Block-chain can help realize collaborative learning in a distributed way without the need for third-party audit. Also, the consensus protocol plays a key technology role. In this article, a trusted consensus scheme for multi-party collaborative learning of edge AI is proposed. To improve system security, reputation-based rights are used to support fast removal of abnormal nodes. With the method of reputation rewards and punishments, the reputations of the nodes participating in the consensus process are scored and recorded on the chain according to their behavior. Nodes who commit malicious actions will be punished by reducing their reputation values. Nodes are rated according to their reputation value, and those with higher reputation rates are acknowledged with more credits and rights in consensus. Since the reputation module is loadable on X-BFT protocols, it is easy to enforce in real applications. The experimental results show that the probability of the attacker being chosen as the leader of XR-BFT is 87.5 percent lower than that of the X-BFT protocols, thus enabling our protocol to achieve trust collaborative learning in edge AI computing with more safety and efficiency.

In this article, we propose and analyze a group of adversarial backdoor attack methods on neural-architecture-search (NAS) enabled edge AI systems in industrial Internet of Things (IIoT) domain. NAS is a new popular way to generate scale-adaptive deep neural networks which can meet the respective requirements of cloud, edge, and terminal AI computing in IIoT domain. However, since most users in NAS-enabled edge side are not the generators of AI models, the deployed edge AI models may have some vulnerabilities such as backdoors. These might pose serious security issues in IIoT. We propose some effective policies to attack such edge AI systems and provide advice about how to defend them. The most significant attack through third-party pretrained NAS in IIoT may occur by backdoor attacks while the third party might introduce vulnerability in the training dataset. The article designs backdoor attack processes to NAS-enabled edge devices to identify NAS’s vulnerability to adversarial trojaning attacks and interpret the backdoor attacks. It shows that the existence of high impact nodes greatly weakens the robustness of the network. A malicious attacker can quickly paralyze the network by only selecting a few high impact nodes. Finally, it provides advice and possible solution on defending the adversarial backdoor attacks to NAS.

Currently, deep-learning-based cognitive automation for decision-making in industrial informatics is a new hot topic in the field of cognitive computing, among which multiobjective architecture optimization is of great difficulty in the research area. When the existing algorithms face multiobjective cognitive model problems, it often takes a lot of time to continuously set different search preference parameters to generate a new search process. This article mainly aims to solve the problem in a multiobjective neural architecture search process, and the key issue is how to adapt user preferences during architectural search. We propose a new algorithm: linear-prefer coevolutionary algorithm. Compared to the original user-constrained method and the Pareto-dominant NSGA-II algorithm, we have faster adaptation time and better quality of adaptation. At the same time, it can respond to user’s needs at a relatively faster pace during the reasoning phase. Based on a large number of comparative test results, our algorithm is superior to the traditional cognitive automation algorithms for the multiobjective problem in search quality.

Edge AI technologies have been used for many Intelligent Transportation Systems, such as road traffic monitor systems. Neural Architecture Search (NAS) is a typcial way to search high-performance models for edge devices with limited computing resources. However, NAS is also vulnerable to adversarial attacks. In this paper, A One-Shot NAS is employed to realize derivative models with different scales. In order to study the relation between adversarial robustness and model scales, a graph-based method is designed to select best sub models generated from One-Shot NAS. Besides, an evaluation method is proposed to assess robustness of deep learning models under various scales of models. Experimental results shows an interesting phenomenon about the correlations between network sizes and model robustness, reducing model parameters will increase model robustness under maximum adversarial attacks, while, increasing model paremters will increase model robustness under minimum adversarial attacks. The phenomenon is analyzed, that is able to help understand the adversarial robustness of models with different scales for edge AI transportation systems.