Pay Less Attention with Lightweight and Dynamic Convolutions

Felix Wu, Angela Fan, Alexei Baevski, Yann Dauphin, Michael Auli

Abstract: Self-attention is a useful mechanism to build generative models for language and images. It determines the importance of context elements by comparing each element to the current time step. In this paper, we show that a very lightweight convolution can perform competitively to the best reported self-attention results. Next, we introduce dynamic convolutions which are simpler and more efficient than self-attention. We predict separate convolution kernels based solely on the current time-step in order to determine the importance of context elements. The number of operations required by this approach scales linearly in the input length, whereas self-attention is quadratic. Experiments on large-scale machine translation, language modeling and abstractive summarization show that dynamic convolutions improve over strong self-attention models. On the WMT'14 English-German test set dynamic convolutions achieve a new state of the art of 29.7 BLEU.

Keywords: Deep learning, sequence to sequence learning, convolutional neural networks, generative models

TL;DR: Dynamic lightweight convolutions are competitive to self-attention on language tasks.

Code: [ pytorch/fairseq](https://github.com/pytorch/fairseq) + [ 3 community implementations](https://paperswithcode.com/paper/?openreview=SkVhlh09tX)

Data: [Billion Word Benchmark](https://paperswithcode.com/dataset/billion-word-benchmark), [CNN/Daily Mail](https://paperswithcode.com/dataset/cnn-daily-mail-1), [WMT 2014](https://paperswithcode.com/dataset/wmt-2014)

Community Implementations: [ 4 code implementations](https://www.catalyzex.com/paper/pay-less-attention-with-lightweight-and/code)