Overview
Explore a thought-provoking lecture on the impact of computational approximations in probabilistic models and machine learning. Delve into the challenges posed by modern large-scale computing and its effects on the expressiveness of probabilistic techniques. Examine how ignoring the influence of computational approximations can undermine both Bayesian principles and the practical utility of inference in real-world applications. Discover a new type of Gaussian Process approximation that provides consistent estimation of the combined posterior, considering both finite observed data and finite computational resources. Learn about the consequences of neglecting computational uncertainty and how implicitly modeling it can improve generalization performance. Gain insights into model selection while accounting for computational factors, and explore an application of these concepts to neurobiological data.
Syllabus
Data is as Data Does: The Influence of Computation on Inference
Taught by
Simons Institute