Trajectory-Guided Steering: Mitigating Spurious Feature Bias in Clinical Classification
arXiv:2610.22618v1 Announce Type: new Abstract: Large language models rely on spurious features in clinical classification: swapping a single gender token in clinical notes alters gender-irrelevant diagnoses in up to 53.5% of cases. Existing inference-time corrections either need counterfactual dat…
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-10-19 04:00 · arXiv cs.CL
Trajectory-Guided Steering: Mitigating Spurious Feature Bias in Clinical Classification