Cebra
CEBRA is a machine learning tool that utilizes non-linear techniques to generate reliable and high-performance latent spaces from joint behavioural and neural data. By mapping behavioural actions to neural activity, CEBRA enhances our understanding of neural dynamics during adaptive behaviours and uncovers underlying correlates of behaviour. This tool is versatile, providing neural latent embeddings for hypothesis testing and discovery-driven analysis. It has been extensively validated on various datasets, including calcium and electrophysiology data, sensory and motor tasks, and simple or complex behaviours across different species.
CEBRA can handle single or multi-session datasets and does not require labels. It excels at mapping and revealing intricate kinematic features, ensuring consistent latent spaces across 2-photon and Neuropixels data, and achieving rapid and accurate decoding of natural movies from the visual cortex. The tool’s code is openly available on GitHub, and a pre-print of its research is accessible on arxiv.org. Neuroscientists seeking to analyze and decode behavioural and neural data to uncover underlying neural representations will find CEBRA to be an invaluable resource.
