Neural Basis of Causal Inference: Representations, Circuits, and Dynamics

To make effective decisions and plan appropriate actions, the brain uses sensory inputs to infer the structure of the environment and the events that take place within it. The process by which the brain infers the objects and events that produce sensory input is known as causal inference. Causal inference is a fundamental computational process that underlies perception in all sensory systems. While behavioral studies have established robust signatures of causal inference, little is known about where and how causal inference takes place in the brain.

With support from an NIH BRAIN Initiative U19 grant, we have formed a highly-collaborative team of theoretical, computational, and experimental neuroscientists with the goal of elucidating fundamental neural mechanisms and circuits that mediate causal inference. We are developing theoretical models that allow us to infer the states and dynamics of hidden (latent) perceptual and motor variables, and which make predictions for how neural population activity is related to these variables. Our experimental studies make use of large-scale neural recordings, as well as chemical and optogenetic approaches for perturbing neural activity, to identify the neural representations of causal inference and the circuits by which our perceptual estimates are updated to maintain consistency with our beliefs about the state of the world. We are particularly interested in exploring how feedback pathways from parietal and prefrontal areas may be involved in updating sensory representations based on causal inferences.

Throughout this project, our goal will be to develop experimental and analytical tools that can be shared with other neuroscientists, as well as a robust data science platform for analyzing and sharing our experimental data.

Projects and Cores

Administrative Core

The Administrative Core will provide the organizational structure and administrative support needed for the CausalityInMotion team to function efficiently and to communicate and disseminate research outcomes. The primary objectives of the Administrative Core will be to provide governance over the research program, to facilitate communication and sharing of ideas, and to establish avenues for sharing information and resources. The Administrative Core will oversee organization of an annual workshop for team members, external advisory board members, and the local scientific communities.

Data Science Core

The Data Science Core will support and enable the team's research efforts by establishing a robust information-technology infrastructure for sharing data and code, by developing a standardized pipeline for data pre-processing, analysis and exchange, by developing novel tools and algorithms for analyzing high-dimensional neural and behavioral data, and by sharing code, data, and other tools with the broader neuroscience community.

Project A: Theory and computational modelling

To study causal inference, we seek to gain insight into the brain's internal model of the world. Since this internal model is only partially accessible through behavior, this project seeks to develop normative models based on the assumption that the brain performs rational operations. This project will develop normative models of the behavioral tasks used in the experimental studies of Projects B and C, and will use these models to generate trial-by-trial predictions of latent variables underlying the causal inference computations. It will further develop quantitative predictions for how these latent variables are represented in neural activity, and how causal manipulations of brain circuits should influence neural and behavioral responses.

Project B: Neural mechanisms of causal inference in trial-based tasks

These experiments will examine the neural basis of perceptual interactions between estimates of object motion, depth, and self-motion. We will explore how making a causal inference regarding object motion in a scene influences behavioral and neural estimates of depth and self-motion velocity. We hypothesize that representations of depth and self-motion velocity will be updated based on current beliefs about independent object motion. We will explore neural correlates of categorical causal inference using array recordings in parietal and prefrontal cortex of macaques, while simultaneously monitoring sensory representations of depth and self-motion velocity using array recordings in visual cortex. We will use optogenetics to identify specific feedback pathways that mediate these effects.

Project C: Neural basis of causal inference in continuous, dynamic navigation tasks

In natural behaviors, our inferences about states of the world evolve dynamically over time and interact with sensory processing through perception-action loops. The goal of this project is to examine causal inference regarding object motion in the context of a continuous "firefly" task, in which subjects navigate through a virtual environment to intercept a briefly-viewed object that may or may not move in the scene. If the object is inferred to be moving, then the subject must extrapolate its movement in order to intercept it. As in Project B, we will study how inferences regarding object motion are represented dynamically in parietal and prefrontal cortex, as well as how sensory representations are dynamically updated in visual cortex based on current beliefs. Experiments will be performed in both macaques and mice, with mouse recordings involving dense sampling throughout the brain.


University of Rochester logo New York University logo Harvard Medical School logo Rice University logo Baylor College of Medicine logo University of Washington logo