5 papers
Mind the Spike: Mechanisms and Brittleness of Visual Massive Activations in Large Vision-Language Models
Jonas Ngnawé, Yann Pequignot, Sabyasachi Sahoo +3
Large vision-language models (LVLMs) inherit massive activations from their text-only bases: spikes where a few fixed hidden channels receive values thousands of times above the ty…
Towards More General Control of Diffusion Models Using Jeffrey Guidance
Raphaël Razafindralambo, Rémy Sun, Frédéric Precioso +2
A key strength of diffusion models lies in their flexibility, since their outputs can be controlled at sampling time through guidance. However, beyond simple cases such as conditio…
Robust Fine-Tuning from Non-Robust Pretrained Models: Mitigating Suboptimal Transfer With Epsilon-Scheduling
Jonas Ngnawé, Maxime Heuillet, Sabyasachi Sahoo +5
Fine-tuning pretrained models is a standard and effective workflow in modern machine learning. However, robust fine-tuning (RFT), which aims to simultaneously achieve adaptation to…
Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers
Jonas Ngnawé, Sabyasachi Sahoo, Yann Pequignot +2
Despite extensive research on adversarial training strategies to improve robustness, the decisions of even the most robust deep learning models can still be quite sensitive to impe…
A Layer Selection Approach to Test Time Adaptation
Sabyasachi Sahoo, Mostafa ElAraby, Jonas Ngnawe +3
Test Time Adaptation (TTA) addresses the problem of distribution shift by adapting a pretrained model to a new domain during inference. When faced with challenging shifts, most met…