AdaptiveMotorControlLab/CEBRA

What it solves

CEBRA addresses the challenge of analyzing high-dimensional time-series recordings, particularly in biology and neuroscience. It helps researchers find consistent, low-dimensional representations (latent spaces) of complex data by using auxiliary variables to guide the learning process, which improves the decoding accuracy of behavioral variables and makes embeddings robust to domain shifts.

How it works

The library implements self-supervised learning algorithms in PyTorch. It uses a patented non-linear clustering method that allows for label-informed time series analysis. By jointly analyzing behavioral and neural data, it produces latent embeddings that capture the underlying variables driving activity and behavior.

Who it’s for

It is primarily designed for researchers in biology and neuroscience who work with high-dimensional recordings, though the tool is applicable to any high-dimensional time-series data.

Highlights

• Self-supervised learning algorithms implemented in PyTorch.
• Support for diverse datasets common in neuroscience and biology.
• Produces embeddings that are robust to domain shifts.
• Improves decoding accuracy of behavioral variables compared to standard supervised learning.