The Center for Cognitive Computation (CCC) invites you to the lecture of Fahd Yazin (University of Edinburgh).
Chair: Mihály Bányai CCC
Title: Will to Predict: How models are created which create our world
Abstract :
Where do predictions come from in the brain? How do these predictions form and shape the models of the world we construct in our mind, driving our subjective experience? Known by various names across fields—mental models, situation models, cognitive maps— I explore the nature of these world models and predictions in sculpting our subjective experience using naturalistic stimuli. Next, I uncover the mechanisms by which these models are formed and deployed for inference using an unsupervised learning task of a probabilistic virtual world.
I show how the Default-Mode Network (DMN) and specifically its Prefrontal sectors segment our environment into abstract, specialized domains - Spatial, Referential and Temporal - through a tripartite architecture, assembling top-down predictions tailored for each domain. This fragments subjective experience, which is unified globally through a multithreaded integration between its prefrontal and parietal core nodes. Computationally, the prefrontal cortex constructs these models by simulating internal data, adaptively adjusting it to match the external reality. Once built, humans compare low-dimensional summaries of these internal replicas to the external sensory information, amounting to rapid model selection.
Formalizing the origin of top-down predictions as being computationally equivalent to predictive inference through Bayesian sampling, I discuss specifically how the prefrontal cortex discovers models and model parameters jointly from the data using internally synthesized data, and generally how the DMN utilizes these to form our subjective experience.