Cebra is a machine learning tool used to compress time series data and reveal hidden structures and variability. It excels in analyzing behavioral neural data and can decode activity in the visual cortex of a mouse brain to reconstruct viewed video.

Cebra can also be applied to rat hippocampus data and 2-photon neuropixels recordings to map the space and uncover complex kinematic features. The tool uses behavioral and neural data in a joint, learnable, and self-supervised manner to produce consistent high-performance latent spaces. It is validated for accuracy and utility across sensory motor tasks and simple and complex behaviors across species.

Cebra also allows single and multi-session datasets to be leveraged for hypothesis testing without the need for labeling. The pre-print paper and software implementation for the algorithm are available on arxiv and Github respectively.