/now

Thesis - Static Malware Detection Across Modalities

Exploring which signals work best for malware detection: text-based static features (e.g., EMBER 2024) vs visual representations (MalNet-style images). Running controlled head-to-head benchmarks and tracking ROC/AUC with an emphasis on lowering false-positive rates. Prototyping ensembles that fuse both modalities to see if they beat either alone. Comparing classic ML (XGBoost/LightGBM) with transformer vision models (ViT) to balance accuracy, robustness, and compute cost.