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

TLDR

Sequence modeling is traditionally linked to recurrent networks, but recent studies demonstrate that convolutional architectures can outperform them on tasks such as audio synthesis and machine translation. The study aims to identify whether convolutional or recurrent models are preferable for a given sequence‑modeling task or dataset. The authors systematically evaluate generic convolutional and recurrent architectures across standard sequence‑modeling benchmarks and provide the code at the cited GitHub repository. Their results show that simple convolutional models outperform LSTMs on diverse tasks, exhibit longer effective memory, and indicate that convolutional networks should be the default starting point for sequence modeling.

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 .