Bootstrap-Conditioned Action Selection with Tabular Foundation Models

2026-08-06

AI Summary

This work addresses practical challenges in contextual bandit deployment, including sparse and biased interaction data, unreliable uncertainty estimation, and severe cold-start issues. The authors propose the BC-ICL strategy, which integrates an arm-context joint conditional architecture with a pretrained tabular foundation model. By leveraging in-context learning and bootstrapped resampling of historical interactions, BC-ICL scores and selects actions under a frozen model setting. This approach effectively enhances statistical information sharing across arms and overcomes the limitations of traditional independent-arm bootstrapping, which often fails in sparse-data regimes. Experimental results demonstrate that BC-ICL significantly outperforms existing baselines on standard benchmarks, achieving particularly strong performance in early rounds and substantially reducing cumulative regret overall.

Abstract

Contextual bandits offer a natural framework for sample-efficient personalization, but practical deployment remains difficult under sparse, biased interaction data, unreliable uncertainty estimates, and severe cold starts. We study whether pre-trained tabular foundation models with in-context learning can be turned into randomized policies for online decision making. We propose BC-ICL (Bootstrap-conditioned action selection using ICL), which at each round draws a bootstrap resample of the interaction history, conditions a frozen pre-trained ICL model on that resample, scores all actions, and selects the action with the highest sampled score. We further introduce an arm-context conditioning architecture that promotes shared statistical strength across actions and helps avoid common bootstrap failure modes of isolated-arm bandits. Empirically, this policy delivers strong early-round regret and regret performance on standard contextual bandit suites, outperforming established baselines under a strict online protocol.

Problem

• contextual bandits
• sparse interaction data
• cold start
• online decision making
• biased data

Innovation

• contextual bandits
• in-context learning
• bootstrap resampling
• tabular foundation models
• online decision making