Abstract:Measurements of neural responses to identically repeated experimental events often exhibit large amounts of variability. This noise is distinct from signal, operationally defined as the average expected response across repeated trials for each given event. Accurately distinguishing signal from noise is important, as each is a target that is worthy of study (many believe noise reflects important aspects of brain function) and it is important not to confuse one for the other. Here, we describe a principled modeling approach in which response measurements are explicitly modeled as the sum of samples from multivariate signal and noise distributions. In our proposed method--termed Generative Modeling of Signal and Noise (GSN)--the signal distribution is estimated by subtracting the estimated noise distribution from the estimated data distribution. Importantly, GSN improves estimates of the signal distribution, but does not provide improved estimates of responses to individual events. We validate GSN using ground-truth simulations and show that it compares favorably with related methods. We also demonstrate the application of GSN to empirical fMRI data to illustrate a simple consequence of GSN: by disentangling signal and noise components in neural responses, GSN denoises principal components analysis and improves estimates of dimensionality. We end by discussing other situations that may benefit from GSN's characterization of signal and noise, such as estimation of noise ceilings for computational models of neural activity. A code toolbox for GSN is provided with both MATLAB and Python implementations.

What problem does this paper attempt to address?

This paper attempts to solve the problem of signal - noise separation in neural response measurement. Specifically, when the same experimental event is repeated, neural response measurements usually show great variability (i.e., noise). This noise is different from the signal, which is defined as the average expected response to each given event in multiple repeated trials. ### Main problems: 1. **Distinguishing between signal and noise**: It is very important to accurately distinguish between signal and noise, because both are research - worthy targets. Many researchers believe that noise reflects important aspects of brain function, so signal and noise should not be confused. 2. **Limitations of existing methods**: The traditional treatment method is to reduce the influence of noise by taking the average of multiple trials, but this cannot completely eliminate the influence of noise. Especially in the case of a small number of trials, the correlation of noise will still remain in the average result. This may lead to inaccurate estimation of signal correlation and affect the results of multivariate analysis, such as principal component analysis, representational similarity analysis and neural dimension analysis. ### Solutions: The paper proposes a new modeling method - Generative Modeling of Signal and Noise (GSN) - for explicitly modeling neural response measurements as the sum of samples from multivariate signal and noise distributions. The core steps of GSN include: - **Estimating data distribution**: Calculate the mean and covariance of the trial - average response to obtain an estimate of the data distribution. - **Estimating noise distribution**: Calculate the residual covariance under each condition and take the average to obtain an estimate of the noise distribution. - **Estimating signal distribution**: Subtract the parameters of the noise distribution from the parameters of the data distribution to obtain an estimate of the signal distribution. ### Verification and application: - **Simulation verification**: The effectiveness of GSN was verified through a series of simulation experiments with known true values, and its performance was compared with other related methods. - **Practical application**: GSN was applied to functional magnetic resonance imaging (fMRI) data in a publicly available natural scene data set, showing how GSN can improve the results of principal component analysis and provide clearer signal estimates and their properties (such as feature spectra and dimensions). In short, this paper aims to more accurately separate the signal and noise in neural responses through the GSN method, thereby improving the understanding of neural activity and the accuracy of analysis.

Disentangling signal and noise in neural responses through generative modeling

Exploiting Noise As a Resource for Computation and Learning in Spiking Neural Networks

Diffusion-Based Generation of Neural Activity from Disentangled Latent Codes

Modeling Latent Neural Dynamics with Gaussian Process Switching Linear Dynamical Systems

Generative modeling and latent space arithmetics predict single-cell perturbation response across cell types, studies and species

SynSigGAN: Generative Adversarial Networks for Synthetic Biomedical Signal Generation

NG-SEM: an effective non-Gaussian structural equation modeling framework for gene regulatory network inference from single-cell RNA-seq data

A generative model of electrophysiological brain responses to stimulation

Generative models for network neuroscience: prospects and promise

Latent Variable Double Gaussian Process Model for Decoding Complex Neural Data

Revisiting the Noise Model of Stochastic Gradient Descent

Data-driven perception of neuron point process with unknown unknowns

Gaussian Process Surrogate Models for Neural Networks

A scalable generative model for dynamical system reconstruction from neuroimaging data

Identifying nonlinear dynamical systems via generative recurrent neural networks with applications to fMRI

Noise and signal: A model-based study of stochastic effects in neural systems

Analyzing neural responses to natural signals: Maximally informative dimensions

Generative Adversarial Network (GAN) for Simulating Electroencephalography

Graph Structured Neural Networks for Perturbation Biology

Linking fast and slow: the case for generative models

Probabilistic Decomposed Linear Dynamical Systems for Robust Discovery of Latent Neural Dynamics