{"id":"arxiv/2606.01655","url":"https://arxiv.org/abs/2606.01655","tags":["math.OC","cs.AI","cs.LG","stat.ML"],"score":0,"title":"MINTS: Minimalist Thompson Sampling","domain":"math.OC","source":"arxiv","authors":["Kaizheng Wang"],"reviews":0,"subject":"math.OC","summary":"The Bayesian paradigm offers principled tools for sequential decision-making under uncertainty, but its reliance on a probabilistic model for all parameters can hinder the incorporation of complex structural constraints. We introduce a minimalist Bayesian framework that places a prior only on the location of the optimum, while eliminating nuisance parameters through profile likelihood. This yields a generalized posterior that naturally accommodates structural constraints. As a direct instantiation, we develop MINimalist Thompson Sampling (MINTS). For multi-armed bandits with mean…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-01","reactions":{}}