Signals 4 · free daily AI digest

Region-Level Black-Box Defense Against Stealthy Embedding-Space Backdoors in CLIP

Paper recorded by Signals 4 on 2026-09-25 in cs.CV. Abstract reproduced from arXiv; link to the original below.

Published 2026-09-25 on arXiv · recorded by Signals 4 on 2026-09-28

Category: cs.CV · 计算机视觉 · first seen 2026-09-28

Abstract

Contrastive Language--Image Pretraining (CLIP) has emerged as a dominant vision backbone due to its strong transferability and zero-shot capabilities. However, recent studies reveal a critical vulnerability: embedding-space backdoor attacks. By poisoning only a tiny fraction of image--text pairs, adversaries can implant stealthy triggers that induce targeted shifts in CLIP's joint embedding space.

Read on arXiv →

#20 most recent of 342 cs.CV papers we have recorded · ↑ newer: How Far Can INRs Go? Cross-Domain Parameter-efficient INR-Based Semant · ↓ older: Structured Reasoning Agentic Framework for Interpretable Critical View
Cite this page: Region-Level Black-Box Defense Against Stealthy Embedding-Space Backdoors in CLIP: the #20 most recent of 342 cs.CV papers we have recorded (as of 2026-09-25). Source: Signals 4 (Signals API) — https://data.jiangzhang.ca/signals4/t/papers/region-level-black-box-defense-against-stealthy-embedding-space-backdoors-in-cli.html
Free to quote with attribution to “Signals 4 (Signals API)”. Machine-readable: papers.json
Related: More cs.CV papers · arXiv signals · All papers · Today in AI
Get 4 AI signals a day by email — free.
Subscribe free → See all plans →
Get 4 AI signals a day by email — free
All models · All repos · By company · Daily editions