Fairness is a critical concern in deep learning, especially in healthcare, where these models influence diagnoses and treatment decisions. Although fairness has been investigated in the vision-only domain, the fairness of medical vision-language (VL) models remains unexplored due to the scarcity of medical VL datasets for studying fairness. To bridge this research gap, we introduce the first fair vision-language medical dataset (Harvard-FairVLMed) that provides detailed demographic attributes, ground-truth labels, and clinical notes to facilitate an in-depth examination of fairness within VL foundation models. Using Harvard-FairVLMed, we conduct a comprehensive fairness analysis of two widely-used VL models (CLIP and BLIP2), pre-trained on both natural and medical domains, across four different protected attributes. Our results highlight significant biases in all VL models, with Asian, Male, Non-Hispanic, and Spanish being the preferred subgroups across the protected attributes of race, gender, ethnicity, and language, respectively. In order to alleviate these biases, we propose FairCLIP an optimal-transport-based approach that achieves a favorable trade-off between performance and fairness by reducing the Sinkhorn distance between the overall sample distribution and the distributions corresponding to each demographic group. As the first VL dataset of its kind, Harvard-FairVLMed holds the potential to catalyze advancements in the development of machine learning models that are both ethically aware and clinically effective. Our dataset and code are available at https://ophai.hms.harvard.edu/datasets/harvard-fairvlmed10k.

In recent years, the use of TLS (Transport Layer Security) protocol to protect communication information has become increasingly popular as users are more aware of network security. However, hackers have also exploited the salient features of the TLS protocol to carry out covert malicious attacks, which threaten the security of network space. Currently, the commonly used traffic detection methods are not always reliable when applied to the problem of encrypted malicious traffic detection due to their limitations. The most significant problem is that these methods do not focus on the key features of encrypted traffic. To address this problem, this study proposes an efficient detection model for encrypted malicious traffic based on transport layer security protocol and a multi-head self-attention mechanism called TLS-MHSA. Firstly, we extract the features of TLS traffic during pre-processing and perform traffic statistics to filter redundant features. Then, we use a multi-head self-attention mechanism to focus on learning key features as well as generate the most important combined features to construct the detection model, thereby detecting the encrypted malicious traffic. Finally, we use a public dataset to verify the effectiveness and efficiency of the TLS-MHSA model, and the experimental results show that the proposed TLS-MHSA model has high precision, recall, F1-measure, AUC-ROC as well as higher stability than seven state-of-the-art detection models.

The growing use of Android software has made mobile devices the main platform for information services such as mobile social media and financial services. Mobile software provides great convenience but also brings challenges to the software community. For example, mobile malware, a malicious software specifically designed to target mobile devices, creates security concerns for the business network and the data stored on it. Therefore, it is becoming more and more important to effectively identify and classify malware. Most of the current malware-classification methods rely on the specific (static/dynamic) behaviour information from Android software for improved malware-detection capability. Nevertheless, these methods cannot detect new types of fraud software due to the limited generalisability. To address these issues, this paper proposes the AMC-FN, i.e. an Android-based malware classification method using feature fusion and natural language processing technologies. The proposed AMC-FN aims to improve the dimension and performance of classification and also some specific functions of natural language processing, i.e. mutual information method, n-gram word segmentation and feature mapping. The AMC-FN framework improves the classification dimensions by leveraging the information from Android APK permission, API calls and realistic network traffic. Moreover, the framework also contains a novel multi-level feature fusion algorithm (MFFA) designed to improve the weighted feature fusion. To obtain better fine granularity and generalisability, the fusion features are used by the optimized SVM (Support Vector Machine) classifier for training. Our experimental measurements and comparisons show the improved performance based on the proposed AMC-FN framework.