English

Better Bounds for the Distributed Experts Problem

Machine Learning 2026-03-11 v1 Data Structures and Algorithms Machine Learning

Abstract

In this paper, we study the distributed experts problem, where nn experts are distributed across ss servers for TT timesteps. The loss of each expert at each time tt is the p\ell_p norm of the vector that consists of the losses of the expert at each of the ss servers at time tt. The goal is to minimize the regret RR, i.e., the loss of the distributed protocol compared to the loss of the best expert, amortized over the all TT times, while using the minimum amount of communication. We give a protocol that achieves regret roughly R1Tpolylog(nsT)R\gtrsim\frac{1}{\sqrt{T}\cdot\text{poly}\log(nsT)}, using O(nR2+sR2)max(s12/p,1)polylog(nsT)\mathcal{O}\left(\frac{n}{R^2}+\frac{s}{R^2}\right)\cdot\max(s^{1-2/p},1)\cdot\text{poly}\log(nsT) bits of communication, which improves on previous work.

Keywords

Cite

@article{arxiv.2603.09168,
  title  = {Better Bounds for the Distributed Experts Problem},
  author = {David P. Woodruff and Samson Zhou},
  journal= {arXiv preprint arXiv:2603.09168},
  year   = {2026}
}
R2 v1 2026-07-01T11:11:37.846Z