PFN-TS: Thompson Sampling for Contextual Bandits via Prior-Data Fitted Networks
Yan Shuo Tan, Kenyon Ng, Ruizhe Deng, Sumetha Loganathan, Qiong Zhang, Bibhas Chakraborty, [0.5em] National University of Singapore, [0.1em] Renmin University of China · 2026
Read PFN-TS: Thompson Sampling for Contextual Bandits via Prior-Data Fitted Networks by Yan Shuo Tan, Kenyon Ng, Ruizhe Deng, Sumetha Loganathan, Qiong Zhang, Bibhas Chakraborty, [0.5em] National University of Singapore, [0.1em] Renmin University of China (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-10137.txt.
Description
Tan and five colleagues adapt prior-data fitted networks for Thompson sampling in contextual bandits, choice problems behind recommendation and adaptive trials. Thompson sampling draws a reward function from a Bayesian posterior each round and acts greedily under that draw. Prior-data fitted networks like TabPFN and TabICL approximate posterior predictive distributions in one forward pass.
The catch: these networks predict noisy future rewards, while Thompson sampling needs uncertainty over the latent mean reward. PFN-TS converts posterior predictives into mean-reward samples through a subsampled predictive central limit theorem. Variance is estimated from a geometric grid of prefixes rather than the full predictive sequence, and cached representations are reused across rounds. The paper proves consistency of the variance estimator and a Bayesian regret bound. PFN-TS earns the top average rank across nonlinear synthetic and OpenML benchmarks, plus the highest policy value in an offline health test.
Catalog record
| CREATOR | Yan Shuo Tan, Kenyon Ng, Ruizhe Deng, Sumetha Loganathan, Qiong Zhang, Bibhas Chakraborty, [0.5em] National University of Singapore, [0.1em] Renmin University of China |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.10137