Abstract:Neurons in the early visual cortex respond selectively to multiple features of visual stimuli, but they respond differently to repeated presentation of the same visual stimulus. Such trial-to-trial response variabilities are often treated as noise and addressed by simple trial-averaging to obtain the stimulus-driven response, though this approach is insufficient to fully remove the response variability. More importantly, there is evidence that response variability may primarily be caused by non-sensory factors, particularly by variations in cortical state. Here we recorded and analyzed neuronal spiking activity in response to natural images from areas 17 and 18 of cats, along with local population neuronal signals; local field potential (LFP) and multi-unit activity (MUA). We used a variability ratio (VR) measure to quantify the variability of neural responses across trials and two cortical state indicative measures; global fluctuation index (GFI) calculated using MUA and synchrony index (SI) calculated using LFP signals. We propose a compact convolutional neural network model with two parallel pathways, to capture the stimulus-driven activity and the cortical state-driven response variabilities. The stimulus-driven pathway contains a spatiotemporal filter, a parametric rectifier and a Gaussian map and the cortical state-driven pathway contains temporal filters for MUA and LFP. The model parameters are fit to best predict the spiking activity of each neuron. Single neurons showed highly varying degrees of trial-to-trial response variability, even when recorded simultaneously. The fitted model performed with a significantly higher accuracy in predicting neural responses compared to a basic model with a stimulus-driven pathway alone. The neurons with higher response variability benefited more from the cortical state-driven pathway compared to less variable neurons. These results provide insights to understanding the possibility of trial-to-trial response variabilities emerging as an effect of cortical state dynamics.

A flow-based latent state generative model of neural population responses to natural images

Generative Modeling of Neural Dynamics via Latent Stochastic Differential Equations

Modeling dynamic neural activity by combining naturalistic video stimuli and stimulus-independent latent factors

Characterizing the nonlinear structure of shared variability in cortical neuron populations using latent variable models

A Tale of Two Latent Flows: Learning Latent Space Normalizing Flow with Short-run Langevin Flow for Approximate Inference

Cortical state contributions to response variability in the early visual cortex: A system identification approach

Enhance Eye Disease Detection using Learnable Probabilistic Discrete Latents in Machine Learning Architectures

Probabilistic Neural Transfer Function Estimation with Bayesian System Identification

A Visually Inspired Computational Model for Recognition of Optic Flow

Conditional Recurrent Flow: Conditional Generation of Longitudinal Samples with Applications to Neuroimaging

Neural Encoding and Decoding with a Flow-based Invertible Generative Model

Flow Matching in Latent Space

Generative random latent features models and statistics of natural images

Linear dynamical neural population models through nonlinear embeddings

Latent Diffusion for Neural Spiking Data

Data-driven perception of neuron point process with unknown unknowns

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

Modeling Latent Neural Dynamics with Gaussian Process Switching Linear Dynamical Systems

Mesoscopic modeling of hidden spiking neurons

Noise Flow: Noise Modeling with Conditional Normalizing Flows

Generative Machine Learning for Detector Response Modeling with a Conditional Normalizing Flow