Neural Collapse in the Intermediate Hidden Layers of Classification Neural Networks
This work provides granular insights into feature propagation in neural networks, but it is incremental as it extends prior studies of NC to intermediate layers.
The paper tackles the problem of understanding Neural Collapse (NC) beyond the final layer by providing the first comprehensive empirical analysis of its emergence in intermediate hidden layers of classification neural networks, finding that NC occurs in most layers with degree correlated to depth and that feature propagation varies with dataset difficulty.
Neural Collapse (NC) gives a precise description of the representations of classes in the final hidden layer of classification neural networks. This description provides insights into how these networks learn features and generalize well when trained past zero training error. However, to date, (NC) has only been studied in the final layer of these networks. In the present paper, we provide the first comprehensive empirical analysis of the emergence of (NC) in the intermediate hidden layers of these classifiers. We examine a variety of network architectures, activations, and datasets, and demonstrate that some degree of (NC) emerges in most of the intermediate hidden layers of the network, where the degree of collapse in any given layer is typically positively correlated with the depth of that layer in the neural network. Moreover, we remark that: (1) almost all of the reduction in intra-class variance in the samples occurs in the shallower layers of the networks, (2) the angular separation between class means increases consistently with hidden layer depth, and (3) simple datasets require only the shallower layers of the networks to fully learn them, whereas more difficult ones require the entire network. Ultimately, these results provide granular insights into the structural propagation of features through classification neural networks.