Generative AI forecasting of malware evolution
Professor: Fabio Di Troia
Description: This research applies machine learning and generative AI to malware detection, with a focus on keeping classifiers effective as malware evolves over time (concept drift) and on detecting previously unseen threats. A current direction, generative forecasting of malware evolution, trains generative models such as diffusion models and fine-tuned language models on time-ordered opcode and API-call sequences to synthesize plausible future malware variants, then uses them to keep detectors accurate before new samples are collected and labeled. The Cal-Bridge Scholar would build time-aware train/test splits on public malware datasets, train and evaluate a generative model that forecasts next-period variants, and measure how much this synthetic augmentation slows classifier decay compared with standard retraining.
Preferred Qualifications: Basic Python, at the level covered in the FAST Workshop, is the only real requirement.
