Learnable latent embeddings for joint behavioural and neural analysishttps://cebra.ai/

Learnable latent embeddings for joint behavioural and neural analysis

🚀 Unlock the mysteries of behavioral & neural data with CEBRA! 🧠📊 This cutting-edge #machinelearning tool decodes neural activity, maps hidden structures, and reconstructs viewed videos with impressive accuracy. 🎥🔍 Dive into neuroscience breakthroughs with CEBRA now! #AI #Neuroscience

  • CEBRA is a machine-learning method designed for joint analysis of behavioral and neural data, excelling in revealing hidden structures in data variability.
  • It can successfully decode activity from the visual cortex of a mouse brain to reconstruct a viewed video, achieving a median absolute error of 5cm.
  • CEBRA is applied to diverse datasets, such as rat hippocampus and mouse primary visual cortex data, to uncover neural dynamics and behavioral correlations.
  • The method uses a novel encoding approach to produce consistent and high-performance latent spaces by leveraging both behavioral and neural data.
  • CEBRA's consistency metric aids in uncovering meaningful differences and enables efficient decoding, proving useful across sensory and motor tasks, different behaviors, and species.
  • The tool accommodates single and multi-session datasets, supporting hypothesis testing and label-free applications.
  • CEBRA is capable of mapping space, revealing complex kinematic features, and providing rapid, high-accuracy decoding of natural movies from the visual cortex.
  • The paper detailing CEBRA is available on arXiv with the identifier arxiv.org/abs/2204.00673.
  • The official implementation of CEBRA can be found on GitHub for further exploration, updates, and collaborations.