Shortcut Features as Top Eigenfunctions of NTK: A Linear Neural Network Case and More
This addresses the chronic problem of shortcut learning in deep learning, which can lead to poor generalization, but the analysis is incremental as it builds on existing NTK theory.
The paper analyzed shortcut learning in neural networks using the Neural Tangent Kernel framework, finding that shortcut features correspond to top eigenfunctions of the NTK and remain influential even after training, with empirical validation on complex networks like ResNet-18.
One of the chronic problems of deep-learning models is shortcut learning. In a case where the majority of training data are dominated by a certain feature, neural networks prefer to learn such a feature even if the feature is not generalizable outside the training set. Based on the framework of Neural Tangent Kernel (NTK), we analyzed the case of linear neural networks to derive some important properties of shortcut learning. We defined a feature of a neural network as an eigenfunction of NTK. Then, we found that shortcut features correspond to features with larger eigenvalues when the shortcuts stem from the imbalanced number of samples in the clustered distribution. We also showed that the features with larger eigenvalues still have a large influence on the neural network output even after training, due to data variances in the clusters. Such a preference for certain features remains even when a margin of a neural network output is controlled, which shows that the max-margin bias is not the only major reason for shortcut learning. These properties of linear neural networks are empirically extended for more complex neural networks as a two-layer fully-connected ReLU network and a ResNet-18.