Regularizing Neural Networks by Penalizing Confident Output Distributions

TL;DR

This paper investigates regularizing neural networks by penalizing confident, low-entropy output distributions, demonstrating that such regularization improves performance across various tasks without changing existing hyperparameters.

Contribution

It introduces a confidence penalty based on maximum entropy principles, connecting it to label smoothing, and thoroughly evaluates its effectiveness across multiple benchmarks.

Findings

Both label smoothing and confidence penalty improve model performance.

Regularizers work across diverse tasks without hyperparameter tuning.

Penalties lead to more calibrated and robust neural network outputs.

Abstract

We systematically explore regularizing neural networks by penalizing low entropy output distributions. We show that penalizing low entropy output distributions, which has been shown to improve exploration in reinforcement learning, acts as a strong regularizer in supervised learning. Furthermore, we connect a maximum entropy based confidence penalty to label smoothing through the direction of the KL divergence. We exhaustively evaluate the proposed confidence penalty and label smoothing on 6 common benchmarks: image classification (MNIST and Cifar-10), language modeling (Penn Treebank), machine translation (WMT'14 English-to-German), and speech recognition (TIMIT and WSJ). We find that both label smoothing and the confidence penalty improve state-of-the-art models across benchmarks without modifying existing hyperparameters, suggesting the wide applicability of these regularizers.