Best-action queries yield Õ(min{T/k, √(T-k)}) regret for i.i.d. stochastic rewards but only Ω(√(T-k)) regret for correlated stochastic or adversarial rewards in the bandit-feedback model.
neurips.cc/paper_files/paper/2017/file/22b1f2e0983160db6f7bb9f62f4dbb39-Paper.pdf
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Multi-Armed Bandits With Best-Action Queries
Best-action queries yield Õ(min{T/k, √(T-k)}) regret for i.i.d. stochastic rewards but only Ω(√(T-k)) regret for correlated stochastic or adversarial rewards in the bandit-feedback model.