The emergence of Vision Language Models (VLMs) has brought unprecedented advances in understanding multi-modal information. The combination of textual and visual semantics in VLMs is highly complex and diverse, making the safety alignment of these models challenging. Furthermore, due to the limited study on the safety alignment of VLMs, there is a lack of large-scale, high-quality datasets. To address these limitations, we propose a Safety Preference Alignment dataset for Vision Language Models named SPA-VL. In terms of breadth, SPA-VL covers 6 harmfulness domains, 13 categories, and 53 subcategories, and contains 100,788 samples of the quadruple (question, image, chosen response, rejected response). In terms of depth, the responses are collected from 12 open-source (e.g., QwenVL) and closed-source (e.g., Gemini) VLMs to ensure diversity. The construction of preference data is fully automated, and the experimental results indicate that models trained with alignment techniques on the SPA-VL dataset exhibit substantial improvements in harmlessness and helpfulness while maintaining core capabilities. SPA-VL, as a large-scale, high-quality, and diverse dataset, represents a significant milestone in ensuring that VLMs achieve both harmlessness and helpfulness.

With a growing security threat in wireless communication networks, a promising method for secure next-generation networks is a zero-trust framework focusing on authentication schemes. How to analyze the risks involved in authentication is a challenge. This study quantifies authentication risks within the zero-trust framework and introduces a privacy domain prevention-control theory. The theory encompasses dynamic privacy risk assessment, intelligent risk classification, and automated selection of privacy protection schemes. First, a dynamic privacy risk assessment method, based on physical entity relationships, is proposed to evaluate all privacy risks. Second, a five-category risk classification method is designed to categorize privacy risks, facilitating the selection of prevention-control schemes, with its rationality mathematically validated. Additionally, an Analytical Hierarchy Process (AHP)-based method is introduced to guide the optimal selection of prevention-control schemes for various scenarios. Finally, the practical application of the theory in medicine multi-modal computing scene of wireless body area networks demonstrates its effectiveness. The experimental results also show the superiority and feasibility of the proposed methods.

Epistasis Detection (ED) was widely used for identifying potential risk disease variants in the human genome. A statistically meaningful ED typically requires a more extensive dataset to detect complex disease-associated Single Nucleotide Polymorphisms (SNPs), but a single institution generally possesses limited genome data. Thus, it is necessary to collect multi-institutional genome data to carry out research together. However, concerns regarding privacy and trustworthiness impede the sharing of massive genome data. Therefore, this study proposes a novel federated ED framework with the sequence perturbation privacy-preserving method to address the limitation of distributed data sharing (FedED-SegNAS). Firstly, to address the lack of interpretability in deep learning models, integrate fuzzy logic into Convolutional Neural Networks (CNNs), promoting the capabilities of CNN to represent the ambiguities of genomic data with high interpretability and reasonable accuracy. Secondly, consider using the Neural Architecture Search (NAS) method to optimize the federated neural architecture. Specifically, selecting the Particle Swarm Optimization (PSO) algorithm to automatically search the optimal neural architecture at different stages in federated learning based on adaptive multi-objectives decreases the communication cost and improves communication efficiency. Furthermore, to ensure the security of the parameter transfer process, design the sequence perturbation privacy-preserving method, grouping the upload and download parameters of federated learning and randomly perturbing the group number so that the attacker cannot obtain the corresponding result between the group number and parameters. Its rationality and security have been proven. The experiments conducted on a range of datasets demonstrate the superiority of the framework over state-of-the-art epistasis detection methods. FedED-SegNAS can reduce network complexity while protecting genome data security.

Zaibin Zhang, Yongting Zhang, Lijun Li, Hongzhi Gao, Lijun Wang, Huchuan Lu, Feng Zhao, Yu Qiao, Jing Shao. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Identifying high-order Single Nucleotide Polymorphism (SNP) interactions of additive genetic model is crucial for detecting complex disease gene-type and predicting pathogenic genes of various disorders. We present a novel framework for high-order gene interactions detection, not directly identifying individual site, but based on Deep Learning (DL) method with Differential Privacy (DP), termed as Deep-DPGI. Firstly, integrate loss functions including cross-entropy and focal loss function to train the model parameters that minimize the value of loss. Secondly, use the layer-wise relevance analysis method to measure relevance difference between neurons weight and outputting results. Deep-DPGI disturbs neuron weight by adaptive noising mechanism, protecting the safety of high-order gene interactions and balancing the privacy and utility. Specifically, more noise is added to gradients of neurons that is less relevance with the outputs, less noise to gradients that more relevance. Finally, Experiments on simulated and real datasets demonstrate that Deep-DPGI not only improve the power of high-order gene interactions detection in with marginal and without marginal effect of complex disease models, but also prevent the disclosure of sensitive information effectively.

The fast development of neuromorphic hardwares promotes Spiking Neural Networks (SNNs) to a thrilling research avenue. Current SNNs, though much efficient, are less effective compared with leading Artificial Neural Networks (ANNs) especially in supervised learning tasks. Recent efforts further demonstrate the potential of SNNs in supervised learning by introducing approximated backpropagation (BP) methods. To deal with the non-differentiable spike function in SNNs, these BP methods utilize information from the spatio-temporal domain to adjust the model parameters. With the increasing of time window and network size, the computational complexity of spatio-temporal backpropagation augments dramatically. In this paper, we propose a new backpropagation method for SNNs based on the accumulated spiking flow (ASF), i.e. ASF-BP. In the proposed ASF-BP method, updating parameters does not rely on the spike train of spiking neurons but leverage accumulated inputs and outputs of spiking neurons over the time window, which reduces the BP complexity significantly. We further present an adaptive linear estimation model to approach the dynamic characteristics of spiking neurons statistically. Experimental results demonstrate that with our proposed ASF-BP method, light-weight convolutional SNNs achieve superior performances compared with other spike-based BP methods on both non-neuromorphic (MNIST, CIFAR10) and neuromorphic (CIFAR10-DVS) datasets. The code is available at https://github.com/neural-lab/ASF-BP.

The fast movement of vehicles and the intermittent connections affect the transmission efficiency in vehicular networks. Considering the regularity of bus routes and departure interval, we utilize buses to improve transmission reliability. This paper focuses on data forwarding at intersections in urban bus ad hoc networks. The traffic information is collected and aggregated by the cluster heads on road segments, and the bus density and the road connectivity are computed accordingly. Then these two parameters as well as the bus line coverage and the path distance are combined to calculate the forwarding priority of a direction at the intersection. Finally, the direction with the highest priority is selected to forward data. Since the direction selection utilizes the real-time traffic information, the feature of bus lines and the road map, the data transmission is enhanced. The simulation results with real road map and bus lines show that our scheme achieves a high delivery ratio and a short delivery latency.