Regularizing Neural Networks via Stochastic Branch Layers

TL;DR

This paper presents StochasticBranch, a regularization method for neural networks that decomposes layers into multiple branches, enabling exploration of diverse parameter configurations during training without increasing inference complexity.

Contribution

Introduces a stochastic regularization technique that decomposes layers into multiple branches, improving generalization without adding inference complexity.

Findings

Effective regularization of neural networks.

Improves generalization performance.

Compatible with existing regularization methods.

Abstract

We introduce a novel stochastic regularization technique for deep neural networks, which decomposes a layer into multiple branches with different parameters and merges stochastically sampled combinations of the outputs from the branches during training. Since the factorized branches can collapse into a single branch through a linear operation, inference requires no additional complexity compared to the ordinary layers. The proposed regularization method, referred to as StochasticBranch, is applicable to any linear layers such as fully-connected or convolution layers. The proposed regularizer allows the model to explore diverse regions of the model parameter space via multiple combinations of branches to find better local minima. An extensive set of experiments shows that our method effectively regularizes networks and further improves the generalization performance when used together with other existing regularization techniques.