ShapKO: Shapley-Adaptive Modality Knockout for Robust Multimodal Learning

2026-07-10

AI Summary

This study addresses the limited robustness of multimodal medical models in real-world clinical settings, where performance often degrades due to missing or dominant modalities. To mitigate this issue, the authors propose a dynamic training strategy that, for the first time, incorporates Shapley values into a modality dropout mechanism. By periodically evaluating the validation utility of modality subsets, the method dynamically estimates each modality’s importance and adaptively adjusts its dropout probability. This approach effectively suppresses overreliance on dominant modalities while encouraging the learning of complementary representations—all without altering the underlying model architecture. Experimental results across three medical benchmark datasets demonstrate significant performance improvements under modality-missing conditions and yield interpretable trajectories of modality dropout decisions.

Abstract

Multimodal medical models often degrade when inputs are missing, a common scenario in real-world clinical workflows. Separately, even when all modalities are present, modality dominance is observed during training, where optimization over-relies on a highly predictive modality and undertrains complementary sources, resulting in poor robustness under partial availability. While training-time modality knockout improves missing-modality robustness, existing approaches use static masking rates that cannot adapt to evolving modality utility during training. We introduce ShapKO (Shapley-Adaptive Modality Knockout), a dynamic training strategy that learns modality-specific knockout probabilities based on validation utility. ShapKO periodically evaluates performance across modality subsets, estimates modality importance via Shapley values, and updates masking probabilities to suppress dominant modalities more frequently. This adaptive process promotes complementary representations, while requiring no architectural modifications. We evaluate ShapKO on three datasets covering multitask clinical classification, survival prediction, and cancer detection. ShapKO consistently improves performance under modality absence and yields interpretable trajectories of learned masking behavior. Code is available at: https://github.com/sumona00/ShapKO

Problem

• multimodal learning
• modality missing
• modality dominance
• robustness
• medical AI

Innovation

• Shapley values
• modality knockout
• multimodal learning
• adaptive masking
• robustness