{
  "id": 7397,
  "url": "https://arxiv.org/abs/2603.10397v1",
  "title": "On the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD",
  "summary": "One crucial factor behind the success of deep learning lies in the implicit bias induced by noise inherent in gradient-based training algorithms. Motivated by empirical observations that training with noisy labels improves model generalization, we delve into the underlying mechanisms behind stochastic gradient descent (SGD) with label noise. Focusing on a two-layer over-parameterized linear network, we analyze the learning dynamics of label noise SGD, unveiling a two-phase learning behavior. In ",
  "authors": "Tongcheng Zhang, Zhanpeng Zhou, Mingze Wang, Andi Han, Wei Huang, Taiji Suzuki et al.",
  "category": "research",
  "topics": "bias-fairness",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-11T04:26:54.000Z",
  "fetched_at": "2026-07-14T16:33:12.390Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/7397",
  "original_url": "https://arxiv.org/abs/2603.10397v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}