TS-Insight: Visualizing Thompson Sampling for Verification and XAI

2025-07-26

AI Summary

Thompson Sampling (TS) and its variants suffer from poor interpretability, debugging difficulty, and low trustworthiness in active learning due to their stochastic decision-making. To address this, we propose TS-Insight—the first visualization analytics framework designed specifically for interpreting TS decision processes. TS-Insight systematically integrates three core components: Bayesian posterior distribution evolution, arm evidence count dynamics, and sample trajectory logging—unified via coordinated, interactive multi-view visualizations that trace the exploration-exploitation trade-off end-to-end. Technically, it combines probabilistic visualization, real-time trajectory replay, and auditable logging to substantially enhance algorithmic transparency and reproducibility. Empirical evaluation demonstrates that TS-Insight effectively supports developers in high-stakes domains—including healthcare and finance—in algorithm validation, bias diagnosis, and trustworthy deployment. By providing a principled, interpretable infrastructure, TS-Insight advances Bayesian active learning toward accountable and human-centered AI.

Technology Category

• Machine Learning: Active Learning
• Reasoning under Uncertainty: Probabilistic Programming
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Responsible Web: Algorithmic accountability and transparency on the web
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Thompson Sampling (TS) and its variants are powerful Multi-Armed Bandit algorithms used to balance exploration and exploitation strategies in active learning. Yet, their probabilistic nature often turns them into a ``black box'', hindering debugging and trust. We introduce TS-Insight, a visual analytics tool explicitly designed to shed light on the internal decision mechanisms of Thompson Sampling-based algorithms, for model developers. It comprises multiple plots, tracing for each arm the evolving posteriors, evidence counts, and sampling outcomes, enabling the verification, diagnosis, and explainability of exploration/exploitation dynamics. This tool aims at fostering trust and facilitating effective debugging and deployment in complex binary decision-making scenarios especially in sensitive domains requiring interpretable decision-making.

Problem

• Visualizing Thompson Sampling for debugging and trust
• Enhancing explainability of exploration/exploitation dynamics
• Supporting interpretable decision-making in sensitive domains

Innovation

• Visual analytics tool for Thompson Sampling
• Traces evolving posteriors and evidence counts
• Enhances trust via explainable decision dynamics