Risk-Sensitive Non-Parametric Bandits Match Classical Regret Bounds
Original title: Risk-Sensitive Non-Parametric and Kernel Bandits
Non-parametric CPT bandits incur no additional order-wise regret cost compared to standard bandits under Cumulative Prospect Theory.
The paper studies non-parametric bandits with arm evaluations via Cumulative Prospect Theory instead of expected reward in continuous spaces with arm-dependent noise. The optimistic NPCB algorithm decouples mean estimation from CPT decisions, proving matching regret bounds. Extensions to unknown noise CDFs and kernel bandits retain comparable rates up to log factors.
- CPT regret order matches classical non-parametric bandits
- Proof for unknown noise CDFs adds distributional Eluder dimension cost
- Information-gain bounds for kernels match risk-neutral rates
From the original article
1 min read at arxiv.orgWe study non-parametric bandits in which arms are evaluated using Cumulative Prospect Theory (CPT) rather than expected reward. We consider continuous action spaces with arm-dependent noise and propose Non-Parametric CPT Bandits ($\mathtt{NPCB}$), an optimistic algorithm that decouples mean estimation from CPT-based decision-making.
By Devdan Dey · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.