Latent Variable Double Gaussian Process Model for Decoding Complex Neural Data

TL;DR

The paper tackles decoding complex neural data with limited samples by introducing a Latent Variable Double Gaussian Process (GP) that models both neural activity and stimuli as outputs of two GPs driven by a shared latent representation $\mathbf{X} \in \mathbb{R}^{N\times Q}$. It uses doubly stochastic variational inference with inducing points to compute a tractable Evidence Lower Bound (ELBO) and infer a robust latent manifold, enabling probabilistic predictions and uncertainty quantification. On the VerbMem EEG dataset, the model achieves a test discrimination accuracy of $92\%$ and an average decoding accuracy of $76.4\%$, outperforming XGBoost with a higher F-measure ($0.766$ vs $0.726$). The work demonstrates a scalable, non-parametric framework that reveals latent structure in neural data and supports accurate decoding while providing principled uncertainty estimates, with open-source tooling to foster adoption and further development.

Abstract

Non-parametric models, such as Gaussian Processes (GP), show promising results in the analysis of complex data. Their applications in neuroscience data have recently gained traction. In this research, we introduce a novel neural decoder model built upon GP models. The core idea is that two GPs generate neural data and their associated labels using a set of low-dimensional latent variables. Under this modeling assumption, the latent variables represent the underlying manifold or essential features present in the neural data. When GPs are trained, the latent variable can be inferred from neural data to decode the labels with a high accuracy. We demonstrate an application of this decoder model in a verbal memory experiment dataset and show that the decoder accuracy in predicting stimulus significantly surpasses the state-of-the-art decoder models. The preceding performance of this model highlights the importance of utilizing non-parametric models in the analysis of neuroscience data.

Figures (2)

• Figure 1: Graphical Model Depicting Latent Variable Double GP: The latent variable $\mathbf{X}$ with a Gaussian prior derives both continuous and discrete covariance structures $\mathbf{K}^n_{NN}$ and $\mathbf{K}^s_{NN}$ for $\mathbf{F}^n$ and $\mathbf{F}^s$ value functions. These functions then define the generative model creating the discrete and continuous measurements.
• Figure 2: Neural Features and Inferred $\mathbf{X}$: (A) ARD coefficients, indicating the relevance of each dimension in the input data to the model. (B-C) Scatter plots of latent variables for training (B) and testing (C) phases. Blue dots represent trials with new words, and red dots indicate trials with old words.