AI & Computational Science

Generalized Regret Analysis of Thompson Sampling using Fractional Posteriors

How the science connects

Bayesian inferenceMulti-armed banditThompson sampling

AI Insight

This paper presents a theoretical analysis of alpha-Thompson Sampling, a variant of the classic Thompson Sampling algorithm for multi-armed bandit problems that uses fractional posteriors (alpha-posteriors with alpha between 0 and 1). The authors prove both instance-dependent and instance-independent regret bounds under mild conditions on prior and reward distributions, showing the algorithm achieves logarithmic regret that matches state-of-the-art methods without requiring restrictive assumptions like conjugate priors or closed-form posteriors.


This work provides theoretical guarantees for a more flexible version of Thompson Sampling that can be applied to a broader class of problems. The relaxed requirements on prior distributions and the elimination of conjugacy constraints make the algorithm more practically applicable while maintaining optimal theoretical performance.


Understand the Science

Bayesian inference Concept coming soon Multi-armed bandit Concept coming soon Thompson sampling Concept coming soon

⚠️ Preprint – Noch nicht peer-reviewed

Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.

-cross
Abstract: Thompson sampling (TS) is one of the most popular and earliest algorithms to solve stochastic multi-armed bandit problems. We consider a variant of TS, named $alpha$-TS, where we use a fractional or $alpha$-posterior ($alphain(0,1)$) instead of the standard posterior distribution. To compute an $alpha$-posterior, the likelihood in the definition of the standard posterior is tempered with a factor $alpha$. For $alpha$-TS we obtain both instance-dependent $mathcal{O}left(sum_{k neq i^*} Delta_kleft(frac{log(T)}{C(alpha)Delta_k^2} + frac{1}{2} right)right)$ and instance-independent $mathcal{O}(sqrt{KTlog K})$ frequentist regret bounds under very mild conditions on the prior and reward distributions, where $Delta_k$ is the gap between the true mean rewards of the $k^{th}$ and the best arms, and $C(alpha)$ is a known constant. Both the sub-Gaussian and exponential family models satisfy our general conditions on the reward distribution. Our conditions on the prior distribution can be easily satisfied by a density that is positive, continuous, and bounded. We also establish another instance-dependent regret upper bound that matches (up to constants) to that of improved UCB [Auer and Ortner, 2010]. Our regret analysis carefully adapts and combines recent theoretical developments in the non-asymptotic concentration analysis and Bernstein-von Mises type results for the $alpha$-posterior distribution. Moreover, our analysis does not require additional structural properties such as closed-form posteriors or conjugate priors.

Source: Generalized Regret Analysis of Thompson Sampling using Fractional Posteriors