Kraken: Higher-order EM Side-Channel Attacks on DNNs in Near and Far Field
This addresses model theft risks for ML practitioners by enabling attacks on modern GPU units, representing an incremental advance over prior work on CUDA Cores.
The paper tackles the problem of stealing DNN models by demonstrating parameter extraction from GPU Tensor Cores via near-field side-channel attacks, achieving this for the first time and showing that electromagnetic radiation leaks up to 100 cm away through obstacles.
The multi-million dollar investment required for modern machine learning (ML) has made large ML models a prime target for theft. In response, the field of model stealing has emerged. Attacks based on physical side-channel information have shown that DNN model extraction is feasible, even on CUDA Cores in a GPU. For the first time, our work demonstrates parameter extraction on the specialized GPU's Tensor Core units, most commonly used GPU units nowadays due to their superior performance, via near-field physical side-channel attacks. Previous work targeted only the general-purpose CUDA Cores in the GPU, the functional units that have been part of the GPU since its inception. Our method is tailored to the GPU architecture to accurately estimate energy consumption and derive efficient attacks via Correlation Power Analysis (CPA). Furthermore, we provide an exploratory analysis of hyperparameter and weight leakage from LLMs in far field and demonstrate that the GPU's electromagnetic radiation leaks even 100 cm away through a glass obstacle.