When majority rules, minority loses: bias amplification of gradient descent

Documenti analoghi

• A second-order-like optimizer with adaptive gradient scaling for deep learning
  di: Bolte, Jérôme, et al.
  Pubblicazione: (2024)
• On the numerical reliability of nonsmooth autodiff: a MaxPool case study
  di: Boustany, Ryan
  Pubblicazione: (2024)
• Geometry-induced Regularization in Deep ReLU Neural Networks
  di: Bona-Pellissier, Joachim, et al.
  Pubblicazione: (2024)
• Precise gradient descent training dynamics for finite-width multi-layer neural networks
  di: Han, Qiyang, et al.
  Pubblicazione: (2025)
• Gradient extremals, talwegs, valleys, and directional alignment for generic gradient descent
  di: Bégout, Pascal, et al.
  Pubblicazione: (2026)