← All publications
CVPR-F26 conference 2026

PoM: A Linear-Time Replacement for Attention with the Polynomial Mixer

David Picard ·Nicolas Dufour ·Lucas Degeorge ·Arijit Ghosh ·Davide Allegro ·Tom Ravaud ·Yohann Perron ·Corentin Sautier ·Zeynep Sonat Baltaci ·Fei Meng ·Syrine Kalleli ·Marta López-Rauhut ·Thibaut Loiseau ·Ségolène Albouy ·Raphael Baena ·Elliot Vincent ·Loic Landrieu

CVPR Findings 2026


Abstract

This paper introduces the Polynomial Mixer (PoM), a novel token mixing mechanism with linear complexity that serves as a drop-in replacement for self-attention. PoM aggregates input tokens into a compact representation through a learned polynomial function, from which each token retrieves contextual information. We prove that PoM satisfies the contextual mapping property, ensuring that transformers equipped with PoM remain universal sequence-to-sequence approximators.