Cebra Logo



Uncovering Hidden Structures in Behavioral and Neural Data with CEBRA

Last Updated:

CEBRA is a machine-learning algorithm that compresses time series data to uncover meaningful differences in behavioral and neural activity. It can be used for both hypothesis-driven and discovery-driven analysis of single or multi-session datasets, with consistent and high-performance latent spaces.

CEBRA is a groundbreaking machine-learning algorithm that helps uncover hidden structures in behavioral and neural data. As the ability to record large neural and behavioral data increases, there is a growing interest in modeling neural dynamics during adaptive behaviors to probe neural representations. CEBRA fills the gap by jointly using behavioral and neural data in a supervised or self-supervised manner to produce consistent and high-performance latent spaces.

CEBRA can be used for hypothesis-driven analysis, where consistent latents can be used to uncover meaningful differences, or for discovery-driven analysis, where inferred latents can be used for decoding. The algorithm is accurate and versatile, allowing for the mapping of space and uncovering of complex kinematic features. CEBRA can be used for both calcium and electrophysiology datasets, across sensory and motor tasks, and in simple or complex behaviors across species.

One of the highlights of CEBRA is its ability to decode and map natural movies from visual cortex, producing rapid and high-accuracy results. The algorithm is also label-free, allowing for single and multi-session datasets to be leveraged for hypothesis testing. The consistent latents produced by CEBRA can be used to uncover underlying correlates of behavior, with neural latent embeddings revealing meaningful differences in behavioral actions and neural activity.

Validation of the accuracy of the algorithm has been demonstrated on various datasets, including those collected from rat hippocampus data and mouse primary visual cortex. The median absolute error is as low as 5cm, which is a significant achievement considering the total track length of 160cm. This showcases the efficacy of CEBRA in decoding activity from the visual cortex of the mouse brain to reconstruct a viewed video.

The pre-print of the paper is available on arxiv, and the official implementation of the CEBRA algorithm can be found on GitHub. Follow their Twitter account or subscribe to their mailing list for updates and releases. If interested in collaborations, you can contact the team through email. Cite this paper using the provided BibTeX format.

Overall, CEBRA is a valuable tool for scientists and researchers looking to uncover latent embeddings in their behavioral and neural data. Its accuracy, versatility, and label-free nature make it a top choice for hypothesis-driven or discovery-driven analysis.