NLDisco: A Pipeline for Interpretable Neural Latent Discovery

This tutorial introduces NLDisco (Neural Latent Discovery pipeline), a machine learning approach for discovering interpretable features in high-dimensional neural recordings. The method leverages sparse autoencoders (SAEs) to identify meaningful latent dimensions that correspond to specific behavioural or environmental variables.

What you’ll accomplish:

• Train sparse autoencoders on neural spike data to automatically discover interpretable features
• Use Matryoshka architecture to capture both broad and specific neural patterns simultaneously
• Validate discoveries through an interactive dashboard that visualizes feature-behavior relationships

Code: NLDisco Tutorial Notebook