Shared Canonical Component Removal Beyond Expert Views
Researchers introduce SCCR to isolate clinical-model signal that expert feature views do not capture.
Shared Canonical Component Removal (SCCR) audits what raw-input models learn beyond expert-derived features by deleting only directions strongly shared between raw and expert embeddings. Unlike residualization or concept erasure, it aims to keep residual predictive signal while limiting redundancy with the expert view. Evaluations on synthetic data, coronary artery disease detection, antidepressant treatment dropout, and SkinCon skin-lesion classification show where expert views are insufficient.
- SCCR removes only embedding directions strongly shared with expert views.
- Tests cover CAD detection, antidepressant dropout, and SkinCon skin lesions.
- It recovers residual predictive signal with less expert-view redundancy.
- The method differs from residualization and concept erasure.
From the original article
1 min read at arxiv.orgIn clinical practice, experts often begin by converting raw inputs into a set of expert features before making a decision. While black-box models trained on raw input can outperform models trained on expert views, it remains unclear what predictive signal the raw model exploits beyond what the expert view captures. We introduce an audit framework to quantify this gap using Shared Canonical Component Removal (SCCR).
By Jose Roberto Tello Ayala · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.