Pay Less Attention with Lightweight and Dynamic Convolutions
This work addresses the high computational cost of self-attention for researchers and practitioners in NLP, offering more efficient methods for tasks like machine translation, though it is incremental as it builds on existing convolution and attention paradigms.
The paper tackles the computational inefficiency of self-attention in generative models by proposing lightweight and dynamic convolutions as alternatives, achieving competitive performance and a new state-of-the-art BLEU score of 29.7 on WMT'14 English-German translation.
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.