Cebra

Overview

CEBRA is a machine learning method that can compress time series data to reveal hidden structures and patterns. It is especially effective on behavioral and neural data recorded simultaneously. CEBRA can even decode visual cortex activity in mice to reconstruct videos they were watching.

How Does CEBRA Work?

CEBRA jointly leverages behavioral and neural data in a supervised or self-supervised manner. This produces consistent, high-performance latent spaces that uncover meaningful relationships in the data. CEBRA's embeddings can be used for accurate decoding and hypothesis testing.

Key Features and Benefits

• Works on calcium and electrophysiology data
• Applicable across sensory, motor, and complex behaviors
• Single and multi-session datasets
• Can be used label-free in a self-supervised manner
• Reveals neural dynamics during natural behaviors
• Rapid, accurate decoding of stimuli from neural data
• Consistent spaces across modalities like 2-photon and Neuropixels

Use Cases and Applications

• Neural mapping of spaces: CEBRA can map hippocampal dynamics to physical spaces traversed by rats.
• Recovering kinematics: It can uncover complex kinematic features from neural data during movement.
• Decoding stimuli: CEBRA enables rapid decoding of natural movies from visual cortex activity.
• Hypothesis testing: The consistent latent spaces allow for powerful hypothesis testing on neural representations.

Who is CEBRA For?

• Neuroscientists studying relationships between neural activity and behavior
• Scientists working with electrophysiology and calcium imaging data
• Researchers interested in decoding stimuli and dynamics from neural data
• Experts studying sensory systems, motor control, and hippocampal spatial tuning

Software and Resources

• CEBRA GitHub Repo - Official implementation
• CEBRA Paper - Published in Nature
• CEBRA Preprint - Available on ArXiv
• CEBRA Twitter - Follow for updates
• CEBRA Mailing List - Subscribe for news
• CEBRA Email - Contact the authors