AI Paper Atlas

Research map / Bandits and online learning / Online convex optimization and games

Improved Regret Bounds for Tracking Experts with Memory

James Robinson, Mark Herbster

NeurIPS 2021 · Bandits and online learning › Online convex optimization and games · approach: Online learning with feedback graphs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace the Markov-prior partition specialists update with PoDS-theta / Share-theta (geometric-decay mixing)Tracking experts with memory · Regret bound, n=500000, k=40, T=4000regret boundImproved
Use memory-based PoDS-theta instead of Fixed-Share when comparison sequence has few unique expertsFixed-Share · Regret bound with m unique expertsregret boundImproved
Set theta to 0 when m = k+1 (no useful memory) (ablation)PoDS-theta / Share-theta · Regret boundregret boundNo clear effect
Replace weight-sharing update with relative entropy projection onto dynamic setGeneralized share update · Update cost (1-norm weight change)weight update costImproved
Pay the switching cost once per switch plus a smaller cost per remembered expert, instead of twice per switchPrevious best memory bound · Regret boundregret boundImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy