Brief

Researchers have developed a new benchmark, "Beyond Classification: A Cough Regression Benchmark for Respiratory Acoustic Foundation Models," to evaluate the performance of foundation models in predicting continuous health metrics from cough audio. The benchmark assesses five models across six targets and three datasets, comparing different regression heads. Results indicate that MLP-small heads outperform baseline predictors, and model performance is influenced by dataset size and head capacity, revealing a trade-off. The study also highlights the asymmetric nature of cross-dataset transfer learning, where large, diverse datasets generalize better to smaller clinical populations than vice versa. AI