Noisy Interpolation Learning with Shallow Univariate ReLU Networks

Understand

Understanding how overparameterized neural networks generalize despite perfect interpolation of noisy training data is a fundamental question.

• Mallinar et.
• al.
• 2022 noted that neural networks seem to often exhibit ``tempered overfitting'', wherein the population risk does not converge to the Bayes optimal error, but neither does it approach infinity, yielding non-trivial generalization.