CEBRA

CEBRA is an open-source machine learning method that compresses high-dimensional time series into consistent, interpretable latent embeddings. Neuroscientists use it to relate recorded brain activity to behavior, and it was published in Nature in 2023.

It can learn embeddings guided by behavioral labels or in a discovery mode without them. Published demonstrations include decoding a rat's position from hippocampus recordings, reconstructing video viewed by a mouse from visual cortex activity, and decoding movement in primates.

The Python package comes with documentation, demo notebooks that open in Google Colab and interactive 3D plots of embeddings. Although built for neural recordings, it can be applied to other paired time-series data.