Linear Attention with Global Context: A Multipole Attention Mechanism for Vision and Physics
This addresses the computational inefficiency of Transformers for high-resolution inputs in vision and physics, offering a scalable solution with broad applications.
The paper tackles the quadratic complexity of standard Transformers by introducing the Multipole Attention Neural Operator (MANO), which achieves linear time and memory complexity while rivaling state-of-the-art models like ViT and Swin Transformer in image classification and Darcy flows, reducing runtime and peak memory usage by orders of magnitude.
Transformers have become the de facto standard for a wide range of tasks, from image classification to physics simulations. Despite their impressive performance, the quadratic complexity of standard Transformers in both memory and time with respect to the input length makes them impractical for processing high-resolution inputs. Therefore, several variants have been proposed, the most successful relying on patchification, downsampling, or coarsening techniques, often at the cost of losing the finest-scale details. In this work, we take a different approach. Inspired by state-of-the-art techniques in $n$-body numerical simulations, we cast attention as an interaction problem between grid points. We introduce the Multipole Attention Neural Operator (MANO), which computes attention in a distance-based multiscale fashion. MANO maintains, in each attention head, a global receptive field and achieves linear time and memory complexity with respect to the number of grid points. Empirical results on image classification and Darcy flows demonstrate that MANO rivals state-of-the-art models such as ViT and Swin Transformer, while reducing runtime and peak memory usage by orders of magnitude. We open source our code for reproducibility at https://github.com/AlexColagrande/MANO.