NTK theory extended to neural network classification

Researchers have extended the Neural Tangent Kernel (NTK) theory to classification tasks, previously a limitation to regression losses. They identified conditions, including parameter-space regularization or non-degenerate targets, under which wide neural networks maintain a constant NTK during training for cross-entropy loss. This allows the training process to be accurately approximated by a linearized model, providing an explicit solution characterization via the NTK and relating model uncertainty to Bayesian methods. AI

IMPACT Extends theoretical understanding of neural network training dynamics for classification tasks.

RANK_REASON Academic paper on extending theoretical framework for neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Neural Tangent Kernel
• Sergio Calvo-Ordoñez

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →