Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a distinguished lecture that presents a formal account of insight and learning through active Bayesian inference, delivered by Karl Friston from University College London. Delve into the dual problem of inferring states of the world while simultaneously learning its statistical structure, contrasting with current machine learning trends like deep learning by focusing on how agents learn from minimal ambiguous outcomes to form insight. Examine simulations of abstract rule-learning and approximate Bayesian inference that demonstrate how minimizing expected free energy leads to active sampling of novel contingencies, resulting in epistemic, curiosity-directed behavior that closes explanatory gaps in knowledge about the world's causal structure. Discover how this process reduces ignorance beyond merely resolving uncertainty about known world states, then progress from inference to model selection and structure learning to understand how abductive processes emerge when agents test plausible hypotheses about symmetries in their generative world models. Learn about the resulting Bayesian model reduction that evokes sleep-associated mechanisms and exhibits characteristics of "aha moments," providing insights into artificial curiosity and the fundamental processes underlying learning and discovery.
