On the Over-Memorization During Natural, Robust and Catastrophic Overfitting

TL;DR

This paper investigates over-memorization in deep neural networks, revealing its role in overfitting and proposing a unified framework to mitigate it by disrupting high-confidence pattern memorization.

Contribution

It introduces the concept of over-memorization as a shared behavior in overfitting, and proposes the DOM framework to prevent it by modifying training patterns.

Findings

Over-memorization impairs generalization in DNNs.

Over-memorization causes high-confidence predictions on training patterns.

The DOM framework effectively reduces overfitting across training paradigms.

Abstract

Overfitting negatively impacts the generalization ability of deep neural networks (DNNs) in both natural and adversarial training. Existing methods struggle to consistently address different types of overfitting, typically designing strategies that focus separately on either natural or adversarial patterns. In this work, we adopt a unified perspective by solely focusing on natural patterns to explore different types of overfitting. Specifically, we examine the memorization effect in DNNs and reveal a shared behaviour termed over-memorization, which impairs their generalization capacity. This behaviour manifests as DNNs suddenly becoming high-confidence in predicting certain training patterns and retaining a persistent memory for them. Furthermore, when DNNs over-memorize an adversarial pattern, they tend to simultaneously exhibit high-confidence prediction for the corresponding natural pattern. These findings motivate us to holistically mitigate different types of overfitting by hindering the DNNs from over-memorization training patterns. To this end, we propose a general framework, Distraction Over-Memorization (DOM), which explicitly prevents over-memorization by either removing or augmenting the high-confidence natural patterns. Extensive experiments demonstrate the effectiveness of our proposed method in mitigating overfitting across various training paradigms.