AI Tools & Models HOTFeaturedAll
arXiv cs.AIScore60

Benign Overfitting for Regression with Trained Two-Layer ReLU Networks

Summary

This study analyses why two-layer ReLU networks trained by gradient descent can overfit regression tasks without harming generalization, revealing the conditions for benign overfitting.

Why this matters

Relevant to this topic

Open original ↗

Source: arXiv cs.AI · arxiv.org

Related items