Catastrophic Overfitting is Weight Drift the Training Loss Cannot See
cs.LG, cs.CR
Submitted: 2026-11-04
Updated: 2026-11-04
Comments: 12 pages (including appendix), 7 figures, 6 tables. Code available at https://github.com/fzjcdt/co-weight-drift
Code: https://github.com/fzjcdt/co-weight-drift
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Explaining and Harnessing Adversarial Examples
- Deep Residual Learning for Image Recognition
- FastAT Benchmark: A Comprehensive Framework for Fair Evaluation of Fast Adversarial Training Methods
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis