Cebra is a machine learning tool for compressing time series data and analyzing neural activity.
Cebra is a machine learning tool specifically designed for compressing time series data, enabling the identification of hidden patterns and variations within it. It particularly excels in analyzing neural data related to behavior and has the capability to decode neural activity within the visual cortex of a mouse brain, reconstructing the video that was viewed.
Moreover, Cebra can be utilized for processing rat hippocampus data as well as 2-photon neuropixels recordings to map spatial information and uncover intricate kinematic characteristics. The tool leverages both behavioral and neural data together, employing a joint, learnable, and self-supervised approach to consistently generate highly effective latent representations. Its accuracy and usefulness have been validated across various species, encompassing sensory motor tasks as well as simple and complex behaviors.
Additionally, Cebra offers the advantage of utilizing single and multi-session datasets for hypothesis testing without requiring manual labeling. The algorithm’s pre-print paper can be found on arxiv, while its software implementation is available on Github.
