Attack-detection mechanisms should be able to reliably detect, predict, and understandably explain both known and novel threats, while also being able to cope with evasion techniques (like obfuscation, polymorphism, or stealth low-volume attacks). They should not just be able to detect any critical incident, but also raise alarms when it matters only; the more false alarms are raised, the higher the risk that users will ignore warnings or even disable the detection permanently. Moreover, to provide a strong line of defense, attack detection methods should be complemented by suitable defensive technologies, ideally selected autonomously by the system under attack, to counter whatever threat it encounters. In this research area we pursue this ideal, currently amongst others by focusing on the detection and analysis of modern malware, on the prevention and attribution of DDoS attacks, and on the identification and mitigation of novel system vulnerabilities.
IEEE Symposium on Security and Privacy (S&P) Zepade Attacks: Exploiting Zero-Page Deduplication for Side-Channel Attacks
International Workshop on Trustworthy and Responsible aUtonomous SysTems(TRUST) Can Open-Weight LLMs Predict Test Outcomes from Traces? A Cross-Subject Boundary Study on Tests4Py
International Workshop on Reliable and trustworthy Automated Software Engineering(RASE) How Grammar-Based Synthetic Data Shapes Behavior Learning: An Empirical Study of Input Generation Strategies
Automated Software Engineering Conference (ASE) Automated Table Reproduction via Code Generation
Automated Software Engineering Conference (ASE) Doc2Feat-Bench: Evaluating Documentation-Driven Feature Addition
