New Feature-Space Smoothing enhances deep learning model robustness

Researchers have developed a new framework called Feature-Space Smoothing (FS) to enhance the robustness of deep learning models against adversarial attacks. FS operates at the representation level, using Gaussian randomized smoothing to guarantee a lower bound on the cosine similarity between clean and adversarial features. This method can be extended to certify predictions and includes a Gaussian Smoothness Booster (GSB) module to improve the model's inherent robustness without requiring retraining. Experiments show that FS integration improves robustness and maintains performance on downstream tasks, particularly for Multimodal Large Language Models (MLLMs). AI

IMPACT Enhances the security and reliability of deep learning models, particularly MLLMs, against adversarial attacks.

RANK_REASON The cluster contains a research paper detailing a new method for improving deep learning model robustness. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Feature Cosine Similarity Bound
• Feature-Space Smoothing
• Gaussian randomized smoothing
• Gaussian Smoothness Booster
• Multimodal Large Language Models and Tunings: Vision, Language, Sensors, Audio, and Beyond
• Song Xia
• University of Cape Town Graduate School of Business

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →