Research map / Adversarial robustness and security / Backdoor and poisoning
Backdoor Attacks under Lossy Compression: From Failure to Reactivation and Adaptation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Studies how lossy image compression (JPEG-style bitstream storage) destroys invisible backdoor triggers and proposes ways to reactivate and adapt attacks under compression.
Problem
Invisible backdoor attacks rely on poisoned datasets that are stored and transmitted, but lossy compression into binary bitstreams can neutralize embedded triggers.
Approach
Shows triggers in RGB images can become ineffective after lossy compression, and develops methods to reactivate and adapt backdoor triggers so they survive compression. Details beyond the abstract are not available.
Surprising findings
- Triggers embedded in RGB images can become ineffective once images are lossily compressed into binary bitstreams.
Closest papers by meaning
- Conditional Backdoor Attack via JPEG Compression (AAAI 2024 · similarity 0.90)
- Data Poisoning based Backdoor Attacks to Contrastive Learning (CVPR 2024 · similarity 0.86)
- Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective (ICLR 2024 · similarity 0.86)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.86)
- Universal Backdoor Attacks (ICLR 2024 · similarity 0.85)
- LOTUS: Evasive and Resilient Backdoor Attacks through Sub-Partitioning (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.