Neural basis of causal inference: representations, circuits, and dynamics

点击翻译按钮获取中文摘要

Project Summary The same pattern of neural activity can correspond to multiple events in the world. Signals sweeping across the retina, for instance, might be generated by a moving object or by the animal's self-motion. The brain resolves this ambiguity by inferring what events best explain sensory activity. This process, called causal inference, is a foundation of action-perception loops in all sensory-motor systems. To support adaptive action, neural representations of variables involved in these computations should be internally consistent. Yet little is known about how such internal models arise, evolve, and interact. This proposal focuses on the neural representations, circuits, and dynamics underlying causal inference in perception of object motion and depth during self-motion. Because the relationships among these variables are defined by physics, not arbitrary trained associations, and because they are likely represented by different cortical areas, the project will be able to study how intercortical connections communicate to maintain an internally consistent view of reality. The overall hypothesis is that causal inference involves computations in parietal and/or prefrontal cortex, and the resulting signals are fed back to sensory areas to update neural representations of task-related variables. Project A will use Bayesian modeling to develop the theoretical framework for studying causal inference in traditional trial-based tasks, and then combine this approach with real-time rational control theory to model continuous, dynamic tasks. These models will be used to fit behavioral data and generate quantitative predictions to compare with behavioral and neural responses in Projects B and C. Using trial-based tasks in monkeys, Project B will ask how causal inference modulates neural correlates of flow parsing (in which background motion influences perception of object motion), will examine how sensory representations are updated by causal inference about object motion, and will use chemical and optogenetic inactivation to identify the specific neural pathways that are necessary for such updating of sensory representations. In naturalistic, continuous navigation tasks, Project C will use similar recording and neural manipulation approaches to examine the neural dynamics of causal inference in monkeys, and will map neural correlates of dynamic causal inference across the entire mouse brain in high-density neural recordings. The Data Science Core will formalize procedures for storing and sharing data, and develop a standard data-processing pipeline, while the Administrative Core will coordinate among the team and manage internal and external advisory committees. These comprehensive research efforts are expected to identify direct correlates of causal inference in single neurons and neural populations and determine how the resulting beliefs about states of the world are propagated from decision-making regions back to sensory regions of the brain. Successful completion of this work will illuminate the functional roles of feedback projections and neural coding in sensory areas of the brain, move the field toward naturalistic continuous behavior, and help close the loop between perception and action.