Скачать или смотреть Enabling AI Attribution, Reuse, and Explainability Through Memory Forensic | David Oygenblik

A proprietary AI model's components (e.g., its code and weights) are often inaccessible after deployment, creating a critical gap for security audits and post-incident forensic investigations. This challenge is amplified on edge devices, where systems often employ online learning, causing the in-the-field model to diverge from any version the original developer can provide. This talk will outline novel forensic methodologies from my research addressing this problem by recovering an AI system’s weights and code directly from a device's memory.

David Oygenblik is a fifth-year Ph.D. student advised by Professor Brendan Saltaformaggio in the School of Electrical and Computer Engineering at the Georgia Institute of Technology. His research interests span program analysis, memory forensics, program instrumentation and forensic investigation of AI system failures. In his free time you’ll catch him hiking, weightlifting, doing escape rooms, playing with his cat, or just enjoying the city of Atlanta.