Generative Models
Collection of resources for generative models in BCI research
Synthesizing EEG Signals from Event-Related Potential Paradigms with Conditional Diffusion Models
A conditional diffusion model is trained on the Lee 2019 ERP dataset. This is the first EEG diffusion model to be conditioned on subject, session and label in parallel, which makes it possible to generate new subject, session, and label-specific data. The model was evaluated using a wide variety of metrics, including the Fréchet inception distance, which utilizes a trained EEGNet for feature extraction.
The EMA weights of the diffusion model after 600k training steps and the trained EEGNet are available for download. More information about the training regime for the diffusion model and EEGNet can be found in:
(2024).
Synthesizing EEG Signals from Event-Related Potential Paradigms with Conditional Diffusion Models.
9th Graz Brain-Computer Interface Conference.
Pierre Guetschel
Ph.D. Student
