An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling

TL;DR

This paper systematically compares convolutional and recurrent neural networks for sequence modeling, finding that convolutional architectures often outperform recurrent ones like LSTMs across various tasks and datasets, suggesting a shift in preferred methods.

Contribution

It provides a comprehensive empirical evaluation showing convolutional networks can surpass recurrent networks in sequence modeling, challenging traditional assumptions.

Findings

Convolutional networks outperform LSTMs on multiple sequence tasks.

Convolutional models demonstrate longer effective memory.

Results advocate for using convolutional architectures as a default choice.

Abstract

For most deep learning practitioners, sequence modeling is synonymous with recurrent networks. Yet recent results indicate that convolutional architectures can outperform recurrent networks on tasks such as audio synthesis and machine translation. Given a new sequence modeling task or dataset, which architecture should one use? We conduct a systematic evaluation of generic convolutional and recurrent architectures for sequence modeling. The models are evaluated across a broad range of standard tasks that are commonly used to benchmark recurrent networks. Our results indicate that a simple convolutional architecture outperforms canonical recurrent networks such as LSTMs across a diverse range of tasks and datasets, while demonstrating longer effective memory. We conclude that the common association between sequence modeling and recurrent networks should be reconsidered, and convolutional networks should be regarded as a natural starting point for sequence modeling tasks. To assist related work, we have made code available at http://github.com/locuslab/TCN .