Adding Gradient Noise Improves Learning for Very Deep Networks

TL;DR

Adding gradient noise during training significantly improves the optimization and performance of very deep neural networks, enabling training from poor initialization and reducing error rates.

Contribution

This paper introduces a simple, low-overhead technique of adding gradient noise that enhances training stability and accuracy for deep and complex neural architectures.

Findings

Enables training of 20-layer networks from poor initialization.

Reduces error rate by 72% on a question-answering task.

Doubles the number of accurate models in binary multiplication across multiple restarts.

Abstract

Deep feedforward and recurrent networks have achieved impressive results in many perception and language processing applications. This success is partially attributed to architectural innovations such as convolutional and long short-term memory networks. The main motivation for these architectural innovations is that they capture better domain knowledge, and importantly are easier to optimize than more basic architectures. Recently, more complex architectures such as Neural Turing Machines and Memory Networks have been proposed for tasks including question answering and general computation, creating a new set of optimization challenges. In this paper, we discuss a low-overhead and easy-to-implement technique of adding gradient noise which we find to be surprisingly effective when training these very deep architectures. The technique not only helps to avoid overfitting, but also can result in lower training loss. This method alone allows a fully-connected 20-layer deep network to be trained with standard gradient descent, even starting from a poor initialization. We see consistent improvements for many complex models, including a 72% relative reduction in error rate over a carefully-tuned baseline on a challenging question-answering task, and a doubling of the number of accurate binary multiplication models learned across 7,000 random restarts. We encourage further application of this technique to additional complex modern architectures.