Reproduce “Diffusion-Based Generation of Neural Activity from Disentangled Latent Codes” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jonathan D. McCart, Andrew R. Sedler, Christopher Versteeg, Domenick Mifsud, Mattia Rigotti-Thompson, Chethan Pandarinath

Researchers at Georgia Tech developed GNOCCHI, a diffusion-based generative model that learns disentangled latent representations of neural activity from high-dimensional recordings. GNOCCHI accurately generates neural activity for unseen conditions and reveals interpretable behavioral variables from neural data without requiring explicit behavioral labels.

Key insights

  • Diffusion models, augmented with information-maximizing objectives like MMD, can learn highly disentangled and interpretable latent representations directly from high-dimensional neural time-series data without explicit behavioral labels.
  • Structured and disentangled latent spaces enable effective 'latent navigation,' allowing for precise, controlled generation of neural activity samples corresponding to specific behavioral conditions while minimizing unintended changes in other factors.
  • The unsupervised discovery of disentangled latent codes provides a robust mechanism for inferring underlying behavioral variables from complex neural activity, thereby reducing reliance on pre-defined experimental paradigms.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.