Yuwen Huang

Yuwen Huang

Postdoc

PLASH: Provably Linear-Time Attention with Selective Higher-Order Feature Sketching

Published in International Conference on Machine Learning (ICML), 2026

Authors: Yuwen Huang and Xiang Pan

Proposes PLASH, an attention block with provably linear-time complexity that preserves the standard attention interface using learned representatives and selective higher-order feature sketching.