AI

Aggregate, Don't Adapt: Subject-Level Posterior Aggregation and Transductive Calibration for Cross-Site Parkinsonian Gait Severity

Researchers have developed a system that predicts Parkinsonian gait severity from motion recordings taken at clinical sites unseen during training. The system, which won a recent challenge, uses a combination of techniques to achieve high accuracy, including averaging per-walk posteriors within subject groups and transductive calibration of feature means and decision operating points. The study found that fine-tuning the encoder led to decreased performance in some cases, and
Researchers have developed a system that predicts Parkinsonian gait severity from motion recordings taken at clinical sites unseen during training. The system, which won a recent challenge, uses a combination of techniques to achieve high accuracy, including averaging per-walk posteriors within subject groups and transductive calibration of feature means and decision operating points. The study found that fine-tuning the encoder led to decreased performance in some cases, and that subject-level aggregation was key to achieving good results. --- Why it matters: This matters because it shows how AI can be used to improve diagnosis and treatment of Parkinson's disease by accurately predicting gait severity from motion recordings. Source: https://arxiv.org/abs/2608.20587

This article was originally published at: https://arxiv.org/abs/2608.20587