Uncertainty-Aware Foundation Models for Clinical Data

2026-04-05

AI Summary

Clinical data are inherently sparse, irregular, and modality-dependent, posing significant challenges for conventional foundation models in handling their incompleteness and uncertainty. This work proposes an uncertainty-aware medical foundation model that uniquely incorporates cognitive uncertainty modeling as a core inductive bias. Instead of point embeddings, the model represents patients through distributed latent states and leverages set-valued representations along with cross-modal consistency constraints to learn generalizable, invariant features. The framework integrates multimodal encoders with self-supervised objectives—combining reconstruction, contrastive alignment, and distributional regularization—to achieve substantial performance gains over strong baselines across diverse clinical tasks. Beyond improved predictive accuracy, the approach enhances robustness to missing data and provides better-calibrated uncertainty estimates.

Abstract

Healthcare foundation models have largely followed paradigms from natural language processing and computer vision, emphasizing large scale pretraining and deterministic representations over heterogeneous clinical data. However, clinical observations are inherently incomplete, reflecting sparse, irregular, and modality dependent measurements of an underlying physiologic state. In this work, we propose a framework for uncertainty aware foundation modeling that represents each patient not as a point embedding, but as a distribution over plausible latent states. By learning set valued representations and enforcing consistency across partial views of the same patient, the model captures what is invariantly inferable while explicitly encoding epistemic uncertainty. We integrate this formulation with multimodal encoders and scalable self supervised objectives, combining reconstruction, contrastive alignment, and distributional regularization. Across diverse clinical tasks, our approach improves predictive performance, robustness under missing data, and uncertainty calibration relative to strong baselines. These results suggest that modeling what is not observed rather than only what is constitutes a critical inductive bias for healthcare foundation models.

Problem

• clinical data
• foundation models
• uncertainty
• incomplete observations
• epistemic uncertainty

Innovation

• uncertainty-aware modeling
• distributional representation
• epistemic uncertainty
• multimodal foundation models
• self-supervised learning