Mega: Moving Average Equipped Gated Attention

TL;DR

Mega introduces a simple, theoretically grounded gated attention mechanism with moving average that efficiently models long sequences, improving performance across various sequence modeling tasks.

Contribution

It presents Mega, a novel attention mechanism incorporating moving average for better inductive bias and efficiency, with a linear complexity variant for long sequence modeling.

Findings

Mega outperforms existing models on multiple benchmarks.

The linear variant maintains high accuracy with reduced complexity.

Mega demonstrates versatility across language, image, and speech tasks.

Abstract

The design choices in the Transformer attention mechanism, including weak inductive bias and quadratic computational complexity, have limited its application for modeling long sequences. In this paper, we introduce Mega, a simple, theoretically grounded, single-head gated attention mechanism equipped with (exponential) moving average to incorporate inductive bias of position-aware local dependencies into the position-agnostic attention mechanism. We further propose a variant of Mega that offers linear time and space complexity yet yields only minimal quality loss, by efficiently splitting the whole sequence into multiple chunks with fixed length. Extensive experiments on a wide range of sequence modeling benchmarks, including the Long Range Arena, neural machine translation, auto-regressive language modeling, and image and speech classification, show that Mega achieves significant improvements over other sequence models, including variants of Transformers and recent state space models.