Towards Demystifying the Generalization Behaviors When Neural Collapse Emerges
This work provides a theoretical explanation for an empirical phenomenon in deep learning generalization, which is incremental but addresses a specific gap in understanding NC.
The paper tackles the lack of theoretical understanding of generalization behaviors during Neural Collapse (NC), establishing a connection between cross-entropy minimization and multi-class SVM to derive a margin generalization bound that explains test accuracy improvement even after 100% train accuracy, with experiments verifying these findings.
Neural Collapse (NC) is a well-known phenomenon of deep neural networks in the terminal phase of training (TPT). It is characterized by the collapse of features and classifier into a symmetrical structure, known as simplex equiangular tight frame (ETF). While there have been extensive studies on optimization characteristics showing the global optimality of neural collapse, little research has been done on the generalization behaviors during the occurrence of NC. Particularly, the important phenomenon of generalization improvement during TPT has been remaining in an empirical observation and lacking rigorous theoretical explanation. In this paper, we establish the connection between the minimization of CE and a multi-class SVM during TPT, and then derive a multi-class margin generalization bound, which provides a theoretical explanation for why continuing training can still lead to accuracy improvement on test set, even after the train accuracy has reached 100%. Additionally, our further theoretical results indicate that different alignment between labels and features in a simplex ETF can result in varying degrees of generalization improvement, despite all models reaching NC and demonstrating similar optimization performance on train set. We refer to this newly discovered property as "non-conservative generalization". In experiments, we also provide empirical observations to verify the indications suggested by our theoretical results.