Learning identifiable and interpretable latent models of high-dimensional neural activity using pi-VAE

TL;DR

This paper introduces pi-VAE, a novel neural data modeling approach that enhances interpretability and identifiability by integrating task variables into latent models, validated on synthetic and real neurophysiological data.

Contribution

The paper adapts identifiable variational auto-encoders for neuroscience, incorporating task variables to improve interpretability and data fitting of neural activity models.

Findings

pi-VAE outperforms existing models in data fitting.

Provides new insights into neural coding structures.

Enhances interpretability of high-dimensional neural data.

Abstract

The ability to record activities from hundreds of neurons simultaneously in the brain has placed an increasing demand for developing appropriate statistical techniques to analyze such data. Recently, deep generative models have been proposed to fit neural population responses. While these methods are flexible and expressive, the downside is that they can be difficult to interpret and identify. To address this problem, we propose a method that integrates key ingredients from latent models and traditional neural encoding models. Our method, pi-VAE, is inspired by recent progress on identifiable variational auto-encoder, which we adapt to make appropriate for neuroscience applications. Specifically, we propose to construct latent variable models of neural activity while simultaneously modeling the relation between the latent and task variables (non-neural variables, e.g. sensory, motor, and other externally observable states). The incorporation of task variables results in models that are not only more constrained, but also show qualitative improvements in interpretability and identifiability. We validate pi-VAE using synthetic data, and apply it to analyze neurophysiological datasets from rat hippocampus and macaque motor cortex. We demonstrate that pi-VAE not only fits the data better, but also provides unexpected novel insights into the structure of the neural codes.