5 June 2024 in IEEE Systems Journal
Edge intelligence has recently attracted great interest from industry and academia, and it greatly improves the processing speed at the edge by moving data and artificial intelligence to the edge of the network. However, edge devices have bottlenecks in battery capacity and computing power, making it challenging to perform computing tasks in dynamic and harsh network environments. Especially in disaster scenarios, edge (rescue) devices are more likely to fail due to unreliable wireless communications and scattered rescue requests, which makes it urgent to explore how to provide low-latency, reliable services through edge collaboration. In this article, we investigate the task offloading mechanism in mobile edge computing networks, aiming to ensure fault tolerance and rapid response of computing services in dynamic and harsh scenarios. Specifically, we design a fault-tolerant distributed task offloading scheme, which minimizes task execution time and system energy consumption through the multi-agent proximal policy optimization algorithm. Furthermore, we introduce logarithmic ratio reward functions and action masking to reduce the impact of different task queue lengths while accelerating model convergence. Numerical results show that the proposed algorithm is suitable for service failure scenarios, effectively meeting the reliability requirements of tasks while simultaneously reducing system energy consumption and processing latency.
1 September 2016 in Automatic Control and Computer Sciences
7 December 2017 in IEEE Access
Computing hardware has become an attractive attack surface due to the globalization of semi-conductor design and supply chain, and the wide integration of third party intellectual property (IP) cores. Recently, gate level information flow tracking (GLIFT) has been proposed to monitor the flow of information in secure hardware design by associating data objects with sensitivity labels and tracking the flow of labeled data. GLIFT can be used to model and verify security-related properties such as confidentiality and integrity. However, existing work in this realm only considers binary labels. These are inadequate for understanding simultaneous information flow behaviors and the root source information flows. In this paper, we propose a precise multi-bit GLIFT method to perform simultaneous multi-bit flow tracking for understanding exactly which bits are affecting a data object at the same time. The proposed method provides more detailed insights into simultaneous information flow behaviors and thus allows proof of quantitative information flow data properties. We compare the complexity and verification performance for different information flow models using primitive gates, IWLS benchmarks, several cryptographic cores and Trust-HUB benchmarks. Experimental results have demonstrated that our method can reason about multi-bit information flow behaviors and identify potential security flaws.
26 April 2024 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
As domain-specific training data is recognized as valuable intellectual property, acquiring well-trained weights in convolutional neural networks (CNN) has emerged as a new threat to the neural network design community. To design a CNN accelerator that is resilient to side-channel threats, it is crucial to have an accurate and efficient security-driven framework at the early design stage. However, there is no standard way to perform root-cause analysis on the power side channel that exists in FPGA-based CNN accelerators. Therefore, we build RE-Specter, a framework that facilitates security-driven design space exploration (DSE) across various building components, combination patterns, and parallelism configurations in CNNs. The goal is to fully understand the power side-channel effects resulting from architectural modifications or optimization decisions. We further compare the benchmarks considering precision, resource utilization, and power side-channel leakage. Finally, we experimentally explore the design space of various architectural features. The experimental results show that low-bit precision delivers more secure architectures ( $68.9\times $ among DSPs, $2439\times $ among LUTs) in measurement-to-disclosure (MTD), but mixed-precision strategies are necessary to maintain the model accuracy. For loop optimization, in 16-parallel scenario, accumulator-based architecture outperforms the architecture featuring an adder tree with the improvements of $8.28\times $ in MTD and $1.38\times $ in PST.
16 July 2025 in IEEE Internet of Things Journal
password-authenticated key exchange (PAKE) allows involved participants to establish cryptographically strong keys over insecure channels by using a low-entropy password as credential. SAE as a widely deployed PAKE in Wi-Fi networks, has no tight security proof and requires the use of the hash-to-curve (H2C) function to map password into an elliptic curve point that is challenging to implement securely. To this end, in this article, we present a tightly secure one-round SAE variant, called as TBSAE for Two-Basis Simultaneous Authentication of Equals. TBSAE can work in both multiplicative and elliptic curve additive groups, and it uses two random group elements as the common reference strings to circumvent the use of Hash-to-Group (Curve) function. In the elliptic curve setting, avoiding the use of H2C function will bring ease-of-implementation advantages over its predecessor SAE and some other H2C-based PAKE protocols since H2C is actually not easy to be implemented securely in practice over a commonly used elliptic curve additive group. TBSAE is peer-to-peer, supporting simultaneous initiation in the authenticated key exchange phase, which makes it also suitable for mesh networks besides the traditional client-server setting. It is secure against the tricky off-line dictionary attack, and further supports the intended forward secrecy. Moreover, we provide a tight formal security proof for TBSAE in the random oracle model using the gap computational Diffie–Hellman (GDH) security assumption, which implies that compared to nontightly secure password-AKE (PAKE) protocols, it can achieve the expected security level with relatively small parameters in practice, making it gain more practical performance advantages.