Alternating Synthetic and Real Gradients for Neural Language Modeling

TL;DR

This paper explores a novel approach to training RNNs for language modeling by alternating between synthetic and real gradients, addressing limitations of traditional backpropagation through time.

Contribution

It introduces a method of combining synthetic and real gradients through alternating training, improving neural language modeling performance.

Findings

Alternating synthetic and real gradients enhances training effectiveness.

Periodic warm restarts improve language modeling results.

The approach offers a viable alternative to BPTT for RNNs.

Abstract

Training recurrent neural networks (RNNs) with backpropagation through time (BPTT) has known drawbacks such as being difficult to capture longterm dependencies in sequences. Successful alternatives to BPTT have not yet been discovered. Recently, BP with synthetic gradients by a decoupled neural interface module has been proposed to replace BPTT for training RNNs. On the other hand, it has been shown that the representations learned with synthetic and real gradients are different though they are functionally identical. In this project, we explore ways of combining synthetic and real gradients with application to neural language modeling tasks. Empirically, we demonstrate the effectiveness of alternating training with synthetic and real gradients after periodic warm restarts on language modeling tasks.