0

Understanding Structured Health Data through Interaction-Aware Mixture-of-Experts

We study interaction-aware mixture-of-experts for post-stroke rigidity prediction using multi-level views of structured health records. Despite minimal performance gains, routing attribution reveals systematic importance differences across views, underscoring view construction…

Open
Year
2026
ArXiv
arxiv.org/abs/2607.12255
Hosting
Abstract onlyARXIV-DEFAULT

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text
arxiv.org/abs/2607.12255ARXIV-DEFAULT
TL;DR
Semantic Scholar
Attribution policy →

Abstract

We study interaction-aware mixture-of-experts for post-stroke rigidity prediction using multi-level views of structured health records. Despite minimal performance gains, routing attribution reveals systematic importance differences across views, underscoring view construction as key to interpretability.