: The increasing deployment of deep learning models has raised important concerns regarding data privacy. In particular, Membership Inference Attacks (MIAs) aim to determine whether a specific data sample was used to train a model, potentially exposing sensitive information. To address this issue, we propose Risk Score guided Knowledge Distillation (RS-KD), a defense mechanism that dynamically adapts the distillation process according to a privacy risk score estimated from the output characteristics of a teacher model. This risk estimation enables the identification of samples that are more vulnerable to MIAs, allowing the framework to selectively regulate the uncertainty introduced during distillation. Experimental results on benchmark datasets demonstrate that RS-KD significantly reduces the effectiveness of black-box MIAs while preserving predictive performance, achieving a favorable privacy–utility trade-off.
R. Kassa, K. Adi, Abdelkamel Tari· Proceedings of the 23rd Inte...· 0 citations
: Accurate cyber threat severity prediction is essential for proactive risk management and efficient resource allocation. Traditional approaches mainly rely on the Common Vulnerability Scoring System (CVSS), which inadequately handles threats exploiting multiple vulnerabilities simultaneously. In this paper, we propose a methodology for cyber threat severity prediction. First, we introduce two aggregation strategies, sequential and parallel exploitation, to compute overall threat severity scores from CVSS metrics while considering multiple exploited vulnerabilities. Second, we propose a predictive model based on Large Language Model (LLM) encoders, namely BERT and the cybersecurity-oriented SecureBERT+. The model processes vulnerability descriptions to generate contextual embeddings, applies attention mechanisms to highlight informative features, and uses Long Short-Term Memory (LSTM) layers to capture contextual dependencies before predicting the threat severity score. Experimental results show improved predictive accuracy and lower error rates, particularly with SecureBERT+.
M. Bekhouche, K. Adi· Proceedings of the 23rd Inte...· 0 citations
