Inference in the Brain: Statistics Flowing in Redundant Population Codes

• Published in: Neuron (Cambridge, Mass.) Vol. 94; no. 5; pp. 943 - 953
• Main Authors: Pitkow, Xaq, Angelaki, Dora E.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 07.06.2017; Elsevier Limited
• Subjects: Activity patterns; Algorithms; Animals; Brain; Brain - physiology; Codes; coding; Decision Making; Humans; Hypotheses; inference; Logic; message-passing; Models, Neurological; Nerve Net - physiology; Neural networks; Neurons; Neurosciences; nonlinear; Nonlinear Dynamics; nuisance; Perception - physiology; Population; population code; Probability; redundant; Reviews; Statistical analysis; Statistics; theory; coding; message-passing; population code; inference; nuisance; brain; nonlinear; theory; redundant
• ISSN: 0896-6273; 1097-4199; 1097-4199
• DOI: 10.1016/j.neuron.2017.05.028

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. Pitkow and Angelaki speculate how the brain could perform inference by passing statistics between redundant, overlapping probabilistic population codes. They argue that neuroscience needs behavioral tasks that include uncertainty and nuisance variables to reveal these key computations.