Understanding the double descent curve in Machine Learning

TL;DR

This paper investigates the double descent phenomenon in machine learning, providing a theoretical framework, exploring its implications for model selection, and validating predictions with experimental results.

Contribution

It offers a fundamental theoretical understanding of double descent and its impact on model selection, which was previously lacking.

Findings

Double descent occurs in over-parameterized models.

Theoretical predictions align with experimental results.

Insights into when double descent is expected to happen.

Abstract

The theory of bias-variance used to serve as a guide for model selection when applying Machine Learning algorithms. However, modern practice has shown success with over-parameterized models that were expected to overfit but did not. This led to the proposal of the double descent curve of performance by Belkin et al. Although it seems to describe a real, representative phenomenon, the field is lacking a fundamental theoretical understanding of what is happening, what are the consequences for model selection and when is double descent expected to occur. In this paper we develop a principled understanding of the phenomenon, and sketch answers to these important questions. Furthermore, we report real experimental results that are correctly predicted by our proposed hypothesis.