Does the multisession mode train a single model for all sessions?

[icarosadero]

I have noticed that the multisession mode requires an ID for each experiment, both in the training and transform methods. Does CEBRA train one global representation for all sessions, or does it generate one separate model for each session?

[MMathisLab]

in the Nature paper we show that you train and encoder per dataset, yet it can leverage information across them. In upcoming work, will will show a single unified encoder; depending on the task, either can be beneficial (i.e., if you want to look for differences across animals, then the current way is best; if you want to make a unified encoder for higher perf. our new variant is best).

[icarosadero]

Currently, I am checking if I can 'mash' everything together as a single session model and train with the partial_fit feature. Thanks for the information