Class Interference of Deep Neural Networks

TL;DR

This paper investigates class interference in deep neural networks, revealing it as a major source of generalization errors and proposing methods to analyze and detect it during training.

Contribution

It introduces new concepts and tools such as cross-class tests, class ego directions, and interference models to understand and measure class interference in neural networks.

Findings

Class interference accounts for the largest portion of generalization errors.

Proposed methods effectively detect class interference during training.

Analyzed minima flatness in relation to class interference.

Abstract

Recognizing and telling similar objects apart is even hard for human beings. In this paper, we show that there is a phenomenon of class interference with all deep neural networks. Class interference represents the learning difficulty in data, and it constitutes the largest percentage of generalization errors by deep networks. To understand class interference, we propose cross-class tests, class ego directions and interference models. We show how to use these definitions to study minima flatness and class interference of a trained model. We also show how to detect class interference during training through label dancing pattern and class dancing notes.