🫛 EDAMAME

Foundation models have recently extended beyond natural language and vision to time-series domains, including physiological signals. Yet developing foundation models for electrodermal activity (EDA) remains hindered by the absence of large-scale, curated, openly accessible datasets: few wearable devices feature EDA sensing, and the only large-scale EDA archive to date is proprietary. To address this gap, we compiled EDAMAME (EDA Multi-dataset Archive for Model training and Evaluation), a collection of EDA traces from 29 public datasets comprising approximately 23,000 hours of data from 731 users. The datasets span six scenarios (sleep monitoring, stress/emotion induction, engagement, real-world stress, workplace analysis, and daily living), both lab and in-the-wild settings, and three devices (Empatica E4, Shimmer 2R, and Shimmer3 GSR+). All data are pre-processed in a unified format and resampled to 4 Hz. Using EDAMAME, we trained UME, the first dedicated foundation model for EDA, whose weights and code are also released.

EDAMAME is available in two forms. The open source version contains the 16 datasets released under open licenses, and can be downloaded without restrictions. The full version adds 11 datasets originally shared under individual data-sharing agreements: their authors agreed to re-share them as part of EDAMAME, so researchers only need to sign one single data-sharing agreement to access all of them, instead of contacting each original provider. The remaining two datasets (DREAMT and HHISS) cannot be re-shared, but are freely available from their original providers, and we release the code to process them into the EDAMAME format.