CEBRA is a machine-learning algorithm that compresses high-dimensional time-series data to reveal hidden structures in neural and behavioral recordings.
What it is
CEBRA is a specialized machine-learning algorithm designed to process high-dimensional time-series recordings. By applying contrastive learning and auxiliary variables, it maps complex neural activity and behavioral movements into lower-dimensional latent spaces. Researchers use this method to uncover hidden patterns, decode sensory information, and analyze multi-session biological datasets.
Key features
- Combines self-supervised and supervised learning approaches for flexible data analysis
- Maps high-dimensional neural and behavioral data into consistent latent spaces
- Reconstructs visual stimuli from mouse visual cortex recordings with high accuracy
- Decodes motor trajectories and spatial positioning from rodent and primate data
- Provides open-source implementation with integration for Python and interactive notebooks
Who it's for
This algorithm is built for computational neuroscientists, machine learning researchers, and academic laboratories studying brain dynamics, neural decoding, and animal behavior through large-scale electrophysiology or calcium imaging datasets.
Pricing
CEBRA is available as an open-source project on GitHub under academic terms, while commercial use requires coordination with the EPFL Tech Transfer Office.
Use cases
- Analyze neural activity from brain recordings
