AI Paper Atlas

Research map / Bandits and online learning / Multi-armed bandits

Stochastic Multi-Armed Bandits with Control Variates

Arun Verma, Manjesh Kumar Hanawal

NeurIPS 2021 · Bandits and online learning › Multi-armed bandits · approach: Non-stationary and structured-reward bandits · 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
use control variates in the UCB index (UCB-CV) versus UCB1, UCB-V, EUCBVUCB-CV · Synthetic Instances 1 and 2 (Gaussian)cumulative regretImproved
compare against EUCBV, the closest baseline, which uses arm eliminationEUCBV · Synthetic Instances 1 and 2cumulative regretHurt
use Thompson Sampling on normal rewards and CVsThompson Sampling · Synthetic Instances 1, 2 and 5cumulative regretHurt
increase reward variance (lower correlation between reward and CV), sigma_v^2 in {1.0,1.5,2.0,2.5,3.0} (ablation)UCB-CV · Instance 5cumulative regretHurt
use Batching instead of UCB-CV for non-normal rewards (ablation)UCB-CV extensions · Instances 2, 3, 4cumulative regretHurt
use Jackknifing or Splitting instead of UCB-CV (ablation)UCB-CV extensions · Instance 3 (Gamma, heavy-tailed)cumulative regretImproved
use Jackknifing or Splitting instead of UCB-CV (ablation)UCB-CV extensions · Instances 2 and 4 (Normal, Log-normal)cumulative regretHurt
add approximation error epsilon to the known CV mean (ablation)UCB-CV · Instances 1, 2, 5cumulative regretHurt

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