The Geometry of Grokking: Norm Minimization on the Zero-Loss Manifold 文章

ArXiv CS.AI2026-06-02NEWSen作者: Tiberiu Musat
查看原文 →
关系图谱

The Geometry of Grokking: Norm Minimization on the Zero-Loss Manifold · 相关技术

相关技术

constrained optimizationlatent representation learningDeep Neural Networkszero-loss manifoldGradient Descentweight decayGrokking