Cebra

What is Cebra?

CEBRA is a machine-learning tool that compresses time series data to reveal hidden patterns in how data changes over time. It can analyze behavioral and brain data simultaneously, using non-linear methods to blend behavioral and neural information for a deeper view of brain activity than older linear approaches.

A notable demonstration: CEBRA can determine what a mouse was seeing from visual cortex activity alone and reconstruct the video. It supports a range of data types, including calcium imaging and electrophysiology, across sensory and motor tasks. Its label-free decoding reconstructs complex movement patterns and visual experiences without external labels, and its multi-session analysis finds consistent patterns across datasets to support hypothesis testing. It can also rapidly and accurately decode natural movies from the visual cortex.

Who created Cebra?

CEBRA was developed by Steffen Schneider, Jin Hwa Lee, and Mackenzie Mathis at EPFL.

What is Cebra used for?

• Linking behavior and brain activity: Maps how specific actions relate to neural activity in neuroscience research.
• Modeling brain dynamics: Models how brain activity changes as animals adapt to new behaviors.
• Neural representations: Uncovers how the brain represents information via neural latent embeddings.
• Decoding visual cortex activity: Decodes mouse visual cortex activity, including reconstructing videos a mouse has seen.
• Joint analysis: Produces latent embeddings usable for analyzing behavior and neural data together.
• Decoding natural movies: Identifies viewed movies from visual cortex activity with high speed and accuracy.
• Multi-session analysis: Explores consistent latent spaces across one or many datasets.
• Hypothesis testing: Supports testing scientific hypotheses with label-free decoding of movement patterns and visual experiences.

Who is Cebra for?

CEBRA is aimed at neuroscientists, neuroscience researchers, and data analysts working with behavioral or neural time series data.