English

On Pareto Optimality for Parametric Choice Bandits

Machine Learning 2026-04-27 v4 Machine Learning

Abstract

We study online assortment optimization under stochastic choice when a decision maker simultaneously values cumulative revenue performance and the quality of post-hoc inference on revenue contrasts. We analyze a forced-exploration optimism-in-the-face-of-uncertainty (OFU) scheme that combines two regularized maximum-likelihood estimators: one based on all observations for sequential decision making, and one based only on exploration rounds for inference. Our general theory is developed under predictable score proxies and per-round action-dependent curvature domination. Under these conditions we establish a self-normalized concentration inequality, a likelihood-based ellipsoidal confidence-set theorem, and a regret bound for approximate optimistic actions that explicitly accounts for optimization error. For the multinomial logit (MNL) model we derive explicit score and curvature proxies and show that a balanced spaced singleton-exploration schedule yields realized coordinate coverage, implying regret \Otilde(nT+T/nT)\Otilde(n_T + T/\sqrt{n_T}) and revenue-contrast error \Otilde(1/nT)\Otilde(1/\sqrt{n_T}) up to fixed problem-dependent factors. A hard two-assortment subclass yields a matching lower bound at the product level. Consequently, within the polynomial exploration family nTTαn_T \asymp T^\alpha, the regret and inference rates become \Otilde(Tmax{α,1α/2})\Otilde(T^{\max\{\alpha,1-\alpha/2\}}) and \Otilde(Tα/2)\Otilde(T^{-\alpha/2}), respectively; hence α[2/3,1)\alpha\in[2/3,1) is the rate-wise Pareto-undominated interval and α=2/3\alpha=2/3 is the unique balancing point that minimizes the regret exponent. Finally, for the Exponomial Choice and Nested Logit models we state verifiable sufficient conditions that would instantiate the general framework.

Keywords

Cite

@article{arxiv.2501.19277,
  title  = {On Pareto Optimality for Parametric Choice Bandits},
  author = {Jierui Zuo and Hanzhang Qin},
  journal= {arXiv preprint arXiv:2501.19277},
  year   = {2026}
}
R2 v1 2026-06-28T21:27:57.256Z