WI Brown Bag Seminar with Matthew Ho

Abstract: The central challenge in modern cosmology is to capitalize on the increasingly diverse and rich observations from next-generation surveys. Machine learning (ML) offers powerful tools to harness this data to test physical theories, yet its adoption requires careful attention to robustness and interpretability. In this talk, I will give a technical overview of the statistical foundations of simulation-based inference, including how we design, validate, and interpret neural networks for scientific modeling. I will motivate why these frameworks are necessary to extract the full physical information from modern surveys and how they can be used to build reliable tests of cosmology and astrophysics. I will outline a research plan that standardizes these techniques for modern surveys, prompts the development of novel physical theory, and establishes ML as a cornerstone of experimental cosmology.