Inference in the Brain: Statistics Flowing in Redundant Population Codes

Abstract

It is widely believed that the brain performs approximate probabilistic inference to estimate causal variables in the world from ambiguous sensory data. To understand these computations, we need to analyze how information is represented and transformed by the actions of nonlinear recurrent neural networks. We propose that these probabilistic computations function by a message-passing algorithm operating at the level of redundant neural populations. To explain this framework, we review its underlying concepts, including graphical models, sufficient statistics, and message-passing, and then describe how these concepts could be implemented by recurrently connected probabilistic population codes. The relevant information flow in these networks will be most interpretable at the population level, particularly for redundant neural codes. We therefore outline a general approach to identify the essential features of a neural message-passing algorithm. Finally, we argue that to reveal the most important aspects of these neural computations, we must study large-scale activity patterns during moderately complex, naturalistic behaviors.

A Dynamic Bayesian Observer Model Reveals Origins of Bias in Visual Path Integration

“…In order to act successfully, we must continuously monitor our sensory inputs, gather evidence in favor of potential actions, and make subjectively good decisions in the face of uncertain evidence. Traditional binary-decision tasks lack the temporal richness to shed light on continuous behaviors in demanding environments 1,2 . Here we develop a visuo-motor virtual navigation task with controllable sensory uncertainty, and provide a unified framework to understand how dynamic perceptual information is combined over time.…”

Human confidence judgments reflect reliability-based hierarchical integration of contextual information

“…In this case, the distribution factorizes so that only local computations (marginalization) need to be performed whose results can be passed on as messages. Hence, the graphical structure of the model facilitates inference which may even be implemented with recurrent neural populations 49 .…”

Motor context coordinates visually guided walking inDrosophila

“…In this study, we examined the role of visual feedback in the control of saccades, and in the capacity of the fly to move in a fixed, straight course. We took an approach guided by the saccade-fixation structure of exploratory locomotion, and by task performance (Harris and Wolpert, 1998;Pitkow and Angelaki, 2017;Todorov, 2004) rather than by aiming at a full description of the modes of walking an exploratory fly may take (Berman et al, 2014;Kabra et al, 2013;Katsov et al, 2017). Although flies can initiate miniscule body saccades while moving forward ( Fig.…”

“…This difficulty becomes more apparent when sensory feedback is perturbed, like attempting to walk on a straight line when blindfolded. Thus, an interesting possibility is that straightness performance depends on the interaction of visual feedback and other self-generated sensory and/or internal signals to generate an robust estimate of path deviations (body state), which can be used in the proper context for course control (Britten, 2008;Dickinson et al, 2000;Franklin and Wolpert, 2011;Körding and Wolpert, 2004;Pitkow and Angelaki, 2017;Todorov and Jordan, 2002).…”