CAPT: A Multi-task Continuous Autoregressive Transformer enabling Cross-dataset and Cross-species Transfer for Calcium Population Dynamics
CAPT is a continuous autoregressive transformer designed to learn calcium population dynamics from single-neuron traces. It uses continuous patch tokenization and self-supervised learning to create a robust representation. The model successfully pretrains on mouse data and demonstrates the ability to transfer its learned knowledge across different species (mouse, zebrafish, C. elegans) and labs by only updating small adaptation modules.