A New Foundation Model for EDA Analysis

A new foundation model, named UME, has been developed specifically for the analysis of electrodermal activity (EDA). This model stands out for being trained on EDAMAME, a vast collection of EDA data from 24 public datasets, comprising over 25,000 hours of recordings from 634 users.

Performance and Resources

UME demonstrates promising performance, outperforming baselines in eight out of ten scenarios and achieving comparable results to more general foundation models, but with a 20x lower computational resource requirement. The results also highlight the intrinsic challenges of EDA modeling, stimulating further research in the field.

Availability

To foster research and development in the field, all datasets used, model weights, and source code have been publicly released.