New framework MABT improves adversarial transfer attacks

Researchers have introduced Manifold Anchored Bilevel Transfer (MABT), a novel framework designed to improve adversarial transfer attacks in machine learning. MABT addresses the issue of adversarial trajectories drifting away from the intrinsic data manifold by anchoring them to a shared semantic subspace. This approach uses a manifold-anchoring operator to reduce noise and frames attack generation as a bilevel optimization problem, learning a geometry-aligned initialization. Experiments show MABT enhances transferability across various attack configurations, victim architectures, and defense mechanisms. AI

IMPACT This research could lead to more robust adversarial attack methods, potentially improving the evaluation of AI model defenses.

RANK_REASON The cluster contains a research paper published on arXiv detailing a new framework for adversarial attacks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• cs.CV
• cs.LG
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv Recommender
• Influence Flower
• Manifold Anchored Bilevel Transfer
• ScienceCast

• paper
• other

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