Dropout as a Regularizer of Interaction Effects

Abstract

We examine Dropout through the perspective of interactions. This view provides a symmetry to explain Dropout: given $N$ variables, there are $\left(\genfrac{}{}{0ex}{}{N}{k}\right)$ possible sets of $k$ variables to form an interaction (i.e. $\mathcal{O}(N^k)$); conversely, the probability an interaction of $k$ variables survives Dropout at rate $p$ is $(1-p{)}^k$ (decaying with $k$). These rates effectively cancel, and so Dropout regularizes against higher-order interactions. We prove this perspective analytically and empirically. This perspective of Dropout as a regularizer against interaction effects has several practical implications: (1) higher Dropout rates should be used when we need stronger regularization against spurious high-order interactions, (2) caution should be exercised when interpreting Dropout-based explanations and uncertainty measures, and (3) networks trained with Input Dropout are biased estimators. We also compare Dropout to other regularizers and find that it is difficult to obtain the same selective pressure against high-order interactions.

References in corpus (11)

• Distilling the Knowledge in a Neural Network
• Improving neural networks by preventing co-adaptation of feature detectors
• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• Learning a SAT Solver from Single-Bit Supervision
• Polynomial Regression As an Alternative to Neural Nets
• SGD on Neural Networks Learns Functions of Increasing Complexity
• Dropout Feature Ranking for Deep Learning Models
• On the Implicit Bias of Dropout
• Dropout as a Structured Shrinkage Prior
• Dropout as a Low-Rank Regularizer for Matrix Factorization
• Visualizing the PHATE of Neural Networks