Label noise in federated learning (FL) is a significant detrimental factor that substantially degrades FL performance. Current methods attempt to mitigate this issue by identifying noisy-labeled samples using rudimentary indicators. However, these indicators fail to distinguish between clean and noisy labels in heavily poisoned scenarios. Although global assessment has been introduced to enhance the recognition of noisy labels, existing methods still struggle to prevent performance degradation in FL systems. Therefore, this paper proposes leveraging the global feature distribution (an integration of local distributions) to improve the detection and correction of noisy labels in FL systems. Specifically, we assess the consistency between the category to which samples belong in the global distribution and their labels to detect noisy labels. Subsequently, a temporal dual-view consistency (TDC) mechanism is designed and introduced to correct the detected noisy labels. TDC evaluates label consistency from the perspectives of sample diversity and label continuity, thereby enhancing the reliability of label correction. Extensive experimental results on both synthetic and real-world noisy datasets demonstrate that the proposed method surpasses current SOTA approaches.

Large Language Models (LLMs) generate realistic and coherent text, boosting efficiency and decision-making in various fields. However, their generative capabilities pose risks of intellectual property abuse. Watermarking technology offers a solution for information hiding in LLMs, with large model watermarking gaining attention for its unique methods. A critical challenge is embedding watermarks with minimal impact on text quality while ensuring rapid detection, making it an urgent issue to address. In this paper, to address these challenge, we propose a universal and robust tagged watermark technology for LLMs (TWT-LLM). Firstly, the method of TWT embeds a certain amount of watermark information in the sampling process while generating subsequent word text based on the prompt. Then, to enhance the quality of the generated text, we have proposed a group-based local watermark embedding method, which significantly reduces the impact on text quality. The method involves tagging tokens within each group, only the tokens that have been tagged will embed the watermark information. Moreover, to detect the watermark information in the generated text, we have designed a detection method specifically for this watermark embedding technique. Finally, we conducted experiments using the C4 and HC3 datasets, demonstrating that TWT-LLM achieves a lower False Negative Rate and is lighter compared to state-of-the-art methods.

In the context of the information age, the frequent occurrence of software vulnerabilities has emerged as a critical issue demanding immediate resolution. Traditional vulnerability detection methods have struggled to keep pace with the escalating security demands, while the advent of deep learning technology has introduced novel solutions to software vulnerability detection. Deep learning not only facilitates the automatic extraction of features, thereby reducing the cost of manual intervention, but also demonstrates remarkable advantages across various domains. In recent years, research on vulnerability detection based on deep learning has achieved notable progress, yet it still faces several limitations. This study focuses on C/C++ program vulnerability detection and proposes an enhanced approach, VulSCS, based on secondary slicing. By performing secondary slicing on source code exhibiting vulnerable behaviors, this method extracts code segments with higher representational value, thereby capturing richer vulnerability-related features. Experimental results indicate that, compared to state-of-the-art vulnerability detection tools, VulSCS improves detection accuracy by 3.2% and enhances detection efficiency by approximately threefold. This research offers new perspectives and methodologies for deep learningbased software vulnerability detection.

Abstract The cipher is a lightweight tweakable block cipher introduced at FSE 2019. Its design aims to incorporate countermeasures against Differential Fault Attacks at the algorithmic level. The cipher employs a lightweight and involutory S-box along with a simple linear layer, enabling efficient encryption and decryption operations. In particular, utilizes a straightforward tweakey schedule that generates four 64-bit round tweakeys, which are reused throughout the encryption process. Despite its lightweight design, the resistance of against impossible differential analysis has not been thoroughly evaluated, with limited attention from cryptanalysts in this regard. Hence, this paper presents a comprehensive analysis of specifically targeting its resistance to impossible differential cryptanalysis. By employing an Satisfiability Modulo Theory (SMT) based automatic search tool, we successfully identify both 12-round related-tweak impossible differential distinguishers and 15-round related-tweakey impossible differential distinguishers for , marking the first discovery of such distinguishers for this cipher. Our results indicate that the tweak in enhances the cipher’s flexibility, provided it receives appropriate attention. In addition, we conduct key-recovery attacks on reduced-round , successfully recovering the 128-bit keys for 20-round, 21-round, and 23-round variants. Based on our comprehensive analysis and experimental results, we conclude that demonstrates effective resistance against impossible differential cryptanalysis.

Virtualization enables the popularization of cloud computing on the one hand, and naturally becomes the security base of cloud computing on the other hand. Nowadays, most of the existing researches focus on the security protection of Virtual Machine (VM) which is ensured by the Virtual Machine Monitor (VMM) provided by Cloud Service Provider. Nevertheless, it's easily neglected that the VMM is a potential malware, which may threaten the confidentiality of VM's data without users' awareness. In this paper, we present HyperDomain, a framework implemented with hardware components and a security VM, aiming to guarantee the confidentiality of data on the memory through verification and measurement of VMM's related operations. Besides, in order to ensure the normal operation of HyperDomain, self-protection mechanisms, including secret communication scheme and capability enhancement of security VM, are introduced. The security analysis shows that the inspection of VMM's misbehavior is effective to defend against the attacks to memory data, and to inform the guest VMs about the illegal operation. In addition, the auxiliary HyperDomain self-protection approaches are proved to be valid for eavesdropping and interruption attacks defense.