Discrete stepping and nonlinear ramping dynamics underlie spiking responses of LIP neurons during decision-making

摘要: Neurons in macaque area LIP exhibit gradual ramping their trial-averaged spike responses during sensory decision-making. However, recent work has sparked debate over whether single-trial trains are better described by discrete "stepping" or continuous drift-diffusion ("ramping99) dynamics. Here we address this controversy using powerful model-based analyses of responses. We extended latent dynamical models to incorporate non-Poisson spiking, baseline firing rates, and various nonlinear relationships between the variable rate. Moreover, used advanced model-comparison methods, including cross-validation a fully Bayesian information criterion, evaluate compare models. These revealed that when spiking was incorporated into existing stepping models, majority neurons remained dynamics, even conditioning on evidence level choice. an model with non-zero compressive output nonlinearity accounted for roughly as many model. The dynamics inferred under exhibited high diffusion variance neurons, making them qualitatively different than slowly-evolving findings generalized alternative tasks, suggesting robust fraction each class.