Pre-Training Buys Better Robustness and Uncertainty Estimates (ICML 2019)

“SURE: SUrvey REcipes for building reliable and robust deep networks” (CVPR 2024) & (ECCV 2024 OOD-CV Challenge Winner)

Effective and Robust Adversarial Training Against Data and Label Corruptions

This project quantifies the geometric and structural "break-point" of LLM representations under data corruption. Using Intrinsic Dimension and CKA mapping, it identifies the exact transformer layers where noise destroys semantic alignment and triggers Uncertainty Calibration collapse.

Node-based glitch art application for building image processing pipelines

This application securely performs deletion by overwriting image files with random data, making them unrecoverable. It’s ideal for safely deleting non-sensitive files.