Test-Time View Selection for Multi-Modal Decision Making

Eeshaan Jain, Johann Wenckstern, Benedikt von Querfurth, Charlotte Bunne

Track: Machine learning: computational method and/or computational results

Nature Biotechnology: Yes

Keywords: active acquisition, active learning, reinforcement learning, clinical diagnostics, foundation models

TL;DR: We introduce MAVIS (Multi-modal Active VIew Selection), a reinforcement learning framework that iteratively identifies the most informative diagnostic tests and markers, aiming to maximize accuracy while minimizing uncertainty and experimental costs

Abstract: The clinical routine possess a growing repertoire of diagnostic tests. Clinical foundation models have recently emerged as powerful tools to extracting and integrate diagnostic information from diverse clinical tests, thereby enabling the creation of patient digital twins. However, it remains unclear which diagnostic tests to select and how to design them to ensure foundation models can extract sufficient information for accurate diagnosis. Here, we introduce MAVIS (Multi-modal Active VIew Selection), a reinforcement learning framework that unifies modality selection and feature selection into a single decision process. By leveraging foundation models, MAVIS dynamically determines which diagnostic tests to perform and in what sequence, adapting to individual patient characteristics. Experiments on real-world datasets across multiple clinical tasks demonstrate that MAVIS outperforms conventional approaches in both diagnostic accuracy and uncertainty reduction, while reducing testing costs by over 80\%, suggesting a promising direction for optimizing clinical workflows through intelligent test design and selection.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Eeshaan_Jain1

Format: Yes, the presenting author will definitely attend in person because they attending ICLR for other complementary reasons.

Funding: No, the presenting author of this submission does *not* fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 125