Cebra

Automatically reduces high-dimensional multi-electrode neural recordings into interpretable low-dimensional representations without requiring labeled data or manual feature engineering. Discovers latent neural dynamics that correlate with actual neural variables and behavioral states.

Processes behavioral video data to automatically discover meaningful latent dimensions representing behavioral structure without manual annotation. Extracts interpretable behavioral representations directly from video frames.

Learns task-agnostic representations from calcium imaging recordings of neural activity, automatically discovering meaningful neural population dynamics without requiring manual region-of-interest definition or behavioral labels.

Creates meaningful low-dimensional embeddings from any time series data without requiring task labels or behavioral annotations. Discovers intrinsic structure in temporal data that is independent of specific experimental tasks.

Analyzes high-dimensional neural recordings with minimal preprocessing requirements, automatically handling data normalization and feature discovery without extensive manual data cleaning or feature engineering steps.

Identifies low-dimensional latent variables that correlate with measurable behavioral and neural variables, producing interpretable representations that can be validated against ground truth measurements. Enables researchers to understand what each discovered dimension represents.

Learns representations that can be compared across different experimental conditions, behavioral states, or recording sessions, enabling discovery of how neural and behavioral dynamics change under different circumstances.

Provides free, open-source code for time series analysis that can be integrated into existing research pipelines and extended by researchers. Enables community contributions and customization for specific research needs.

Provides a superior alternative to conventional dimensionality reduction methods like PCA by discovering task-agnostic representations that capture behavioral and neural structure better than linear methods. Outperforms traditional approaches on complex neural and behavioral data.