CCN Seminar with Guest Speaker Amy Christensen (Washington University in St. Louis)

Please join us for a talk by Amy Christensen, Postdoctoral Fellow, Kepecs Lab at Washington University in St Louis. 

Title: Inferring the neural dynamics of decision-making under uncertainty
 
Abstract:
How do rats modify their behavior as a function of environmental and internal uncertainty? My talk will feature two vignettes that address this question from different perspectives:
 
1. The environment contains latent and changing structure not directly knowable from momentary sensory input. However, knowledge of this latent structure can be learned through interaction with the environment, and once learned is useful to infer the true underlying causes of uncertain sensory input (e.g. it is useful for perceptual inference).  To study these perceptual learning and inference processes, in graduate school I developed a visual discrimination task with blocked hidden state-transitions that govern stimulus generation. In this paradigm I could track changes in the activity of single neurons over two distinct timescales, first as rats learned about the existence of the latent block structure (over the course of a week), and then in a single session as a function what block the rat estimates it’s in. My talk will explore the behavioral and neural correlates of these two timescales of learning and inference!
 
2. Estimation of internal uncertainty about previous decisions (e.g. confidence) is an important factor to consider when contemplating future, related decisions. To study the neural dynamics that support this sequential decision-making process, in my postdoctoral work I trained rats to perform a confidence-guided time-investment task. In this task, the amount of time rats are willing to wait for a delayed reward is proportional to their confidence in a previous perceptual decision. From neural spiking activity we first identified functional neuron clusters with distinct temporal profiles, and then used highly constrained recurrent neural networks to infer the equations of a sparse, low-dimensional dynamical system that could generate interpretable single trial, cluster-averaged activity. Our findings reveal a dynamical system underlying a decision process in the orbitofrontal cortex and provide a promising approach for the discovery of dynamical systems algorithms that explain behavior directly from neural data.



Organizer: Center for Computational Neuroscience