Physicists at Harvard University have created a simplified, physics-inspired mathematical model to better understand how neural networks learn without overfitting. The model, based on ridge regression ...