Revisiting Matrix Sketching in Linear Bandits: Achieving Sublinear Regret via Dyadic Block Sketching
Dongxie Wen · Hanyan Yin · Xiao Zhang · Peng Zhao · Lijun Zhang · Zhewei Wei
Abstract
Linear bandits have become a cornerstone of online learning and sequential decision-making, providing solid theoretical foundations for balancing exploration and exploitation. Within this domain, matrix sketching serves as a critical component for achieving computational efficiency, especially when confronting high-dimensional problem instances. The sketch-based approaches reduce per-round complexity from $\Omega(d^2)$ to $O(dl)$, where $d$ is the dimension and $l
Successful Page Load