A pipeline for interpretable neural latent discovery

Jai Bhagat, Anaya Gaelle Pouget, Sara Molas-Medina

Reviewer: ~Jai_Bhagat1

Presenter: ~Jai_Bhagat1

TL;DR: NLDisco adapts sparse dictionary learning from AI interpretability to discover interpretable features in neural recordings, offering an alternative to traditional latent variable models with opaque latent spaces.

Abstract: Mechanistic understanding of the brain requires interpretability of large-scale neuronal computations. Many latent variable model approaches excel at decoding, but produce complex, opaque latent spaces. We address this with NLDisco, a pipeline for interpretable neural latent discovery. Motivated by successful applications of sparse dictionary learning in AI mechanistic interpretability, NLDisco encourages hidden layer neurons in sparse encoder-decoder models to learn interpretable representations. A flexible, user-friendly software package supports interpretable latent discovery across recording modalities and experimental paradigms. We validate the pipeline on a synthetic dataset, demonstrating that it recovers ground-truth features and reveals meaningful representations. We conclude by discussing future development and applications, emphasizing the pipeline's potential to facilitate neuroscientific discovery and clearer insights into neural computations.

Length: short paper (up to 4 pages)

Domain: methods

Format Check: Yes, the presenting author will definitely attend in person because they attending NeurIPS for other complementary reasons.

Author List Check: The author list is correctly ordered and I understand that additions and removals will not be allowed after the abstract submission deadline.

Anonymization Check: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and URLs that point to identifying information.

Submission Number: 72