Augmenting Self-attention with Persistent Memory

TL;DR

This paper introduces a novel transformer variant that replaces the feed-forward layer with persistent memory vectors in self-attention layers, maintaining performance while simplifying the architecture.

Contribution

The authors propose a transformer model that solely uses attention layers augmented with persistent memory, removing the need for feed-forward layers.

Findings

Maintains performance on language modeling benchmarks

Simplifies transformer architecture by removing feed-forward layers

Demonstrates benefits on character and word level tasks

Abstract

Transformer networks have lead to important progress in language modeling and machine translation. These models include two consecutive modules, a feed-forward layer and a self-attention layer. The latter allows the network to capture long term dependencies and are often regarded as the key ingredient in the success of Transformers. Building upon this intuition, we propose a new model that solely consists of attention layers. More precisely, we augment the self-attention layers with persistent memory vectors that play a similar role as the feed-forward layer. Thanks to these vectors, we can remove the feed-forward layer without degrading the performance of a transformer. Our evaluation shows the benefits brought by our model on standard character and word level language modeling benchmarks.