Research map / Optimization / Implicit bias and generalization
A Theoretical Analysis of Backdoor Learning as Simplicity-Biased Optimization Dynamics
Hong Cheng, Qiang Zhou, Sinno Pan
Read the paperCheck your idea against 137,217 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.
Proposes a theoretical account that backdoor learning arises from simplicity-biased optimization dynamics, in which the simple trigger-target association is learned readily under standard training.
Problem
Why standard training dynamics learn backdoor trigger-target associations so effectively while preserving clean performance remains without a unified explanation.
Approach
Frames backdoor learning as a consequence of simplicity bias in optimization dynamics: the trigger is a simple feature that gets learned preferentially, offering a unified theoretical explanation rather than purely empirical study.
Closest papers by meaning
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.91)
- Indelible Backdoors: On the Limits of Post-Training Defenses (ECCV 2026 · similarity 0.90)
- Don't Shift the Trigger: Robust Gradient Ascent for Backdoor Unlearning (ICLR 2026 · similarity 0.90)
- Backdoor Mitigation via Invertible Pruning Masks (NeurIPS 2025 · similarity 0.88)
- REFINE: Inversion-Free Backdoor Defense via Model Reprogramming (ICLR 2025 · similarity 0.88)
- Backdoor Attacks in Token Selection of Attention Mechanism (ICML 2025 · similarity 0.88)
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.
