Controlling Neural Collapse Enhances Out-of-Distribution Detection and Transfer Learning

TL;DR

This paper explores the relationship between Neural Collapse and out-of-distribution detection and transfer learning, proposing methods to control Neural Collapse at different layers to improve both tasks.

Contribution

It introduces a theoretical framework linking Neural Collapse to OOD detection and generalization, and proposes a method to control Neural Collapse at different layers.

Findings

Controlling Neural Collapse improves OOD detection performance.

Reducing Neural Collapse enhances out-of-distribution generalization.

The proposed method outperforms existing approaches across datasets and architectures.

Abstract

Out-of-distribution (OOD) detection and OOD generalization are widely studied in Deep Neural Networks (DNNs), yet their relationship remains poorly understood. We empirically show that the degree of Neural Collapse (NC) in a network layer is inversely related with these objectives: stronger NC improves OOD detection but degrades generalization, while weaker NC enhances generalization at the cost of detection. This trade-off suggests that a single feature space cannot simultaneously achieve both tasks. To address this, we develop a theoretical framework linking NC to OOD detection and generalization. We show that entropy regularization mitigates NC to improve generalization, while a fixed Simplex Equiangular Tight Frame (ETF) projector enforces NC for better detection. Based on these insights, we propose a method to control NC at different DNN layers. In experiments, our method excels at both tasks across OOD datasets and DNN architectures. Code for our experiments is available at: https://yousuf907.github.io/ncoodg