Why Most Optimism Bandit Algorithms Have the Same Regret Analysis: A Simple Unifying Theorem

Fuente: arXiv
Saved in:
Bibliographic Details
Main Author: Krishnamurthy, Vikram
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908725337391104
author Krishnamurthy, Vikram
author_facet Krishnamurthy, Vikram
contents Several optimism-based stochastic bandit algorithms -- including UCB, UCB-V, linear UCB, and finite-arm GP-UCB -- achieve logarithmic regret using proofs that, despite superficial differences, follow essentially the same structure. This note isolates the minimal ingredients behind these analyses: a single high-probability concentration condition on the estimators, after which logarithmic regret follows from two short deterministic lemmas describing radius collapse and optimism-forced deviations. The framework yields unified, near-minimal proofs for these classical algorithms and extends naturally to many contemporary bandit variants.
format Preprint
id arxiv_https___arxiv_org_abs_2512_18409
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Why Most Optimism Bandit Algorithms Have the Same Regret Analysis: A Simple Unifying Theorem
Krishnamurthy, Vikram
Machine Learning
Systems and Control
Several optimism-based stochastic bandit algorithms -- including UCB, UCB-V, linear UCB, and finite-arm GP-UCB -- achieve logarithmic regret using proofs that, despite superficial differences, follow essentially the same structure. This note isolates the minimal ingredients behind these analyses: a single high-probability concentration condition on the estimators, after which logarithmic regret follows from two short deterministic lemmas describing radius collapse and optimism-forced deviations. The framework yields unified, near-minimal proofs for these classical algorithms and extends naturally to many contemporary bandit variants.
title Why Most Optimism Bandit Algorithms Have the Same Regret Analysis: A Simple Unifying Theorem
topic Machine Learning
Systems and Control
url https://arxiv.org/abs/2512.18409