🤖 AI Summary
This study investigates the epoch-wise double descent phenomenon in deep neural networks trained on noisy data and its intrinsic connections to benign overfitting and large activation mechanisms. By training fully connected networks of varying sizes on CIFAR-10 with 30% label noise, the authors integrate loss decomposition, activation pattern tracking, and layer-wise representation analysis into a unified empirical framework, for the first time linking epoch-wise double descent, benign overfitting, and large activations in shallow layers. The findings reveal that models can achieve strong generalization even after fully fitting noisy labels; activations corresponding to noisy and clean samples gradually separate in deeper layers; and large activations emerge in shallow layers—correlated with input structure—whose evolution aligns precisely with the re-generalization process.
📝 Abstract
Deep double descent is one of the key phenomena underlying the generalization capability of deep learning models. In this study, epoch-wise double descent, which is delayed generalization following overfitting, was empirically investigated by focusing on the evolution of internal structures. Fully connected neural networks of three different sizes were trained on the CIFAR-10 dataset with 30% label noise. By decomposing the loss curves into signal contributions from clean and noisy training data, the epoch-wise evolutions of internal signals were analyzed separately. Three main findings were obtained from this analysis. First, the model achieved strong re-generalization on test data even after perfectly fitting noisy training data during the double descent phase, corresponding to a"benign overfitting"state. Second, noisy data were learned after clean data, and as learning progressed, their corresponding internal activations became increasingly separated in outer layers; this enabled the model to overfit only noisy data. Third, a single, very large activation emerged in the shallow layer across all models; this phenomenon is referred as"outliers,""massive activa-tions,"and"super activations"in recent large language models and evolves with re-generalization. The magnitude of large activation correlated with input patterns but not with output patterns. These empirical findings directly link the recent key phenomena of"deep double descent,""benign overfitting,"and"large activation", and support the proposal of a novel scenario for understanding deep double descent.