Abstract:
AI is undoubtedly impressive when applied to text and images. Can it work on physical systems, governed by Partial Differential Equations (PDEs)? Many are interested in the answer to this question, but it is not a trivial one to unpack. What does it even mean to combine machine learning and PDEs? What are the tasks that neural networks can accomplish in this context? In this talk, I aim to give a bird’s-eye overview of the subjects, focusing less on fine architectural details and more on the overall approach. We will explore different paradigms for combining ML and PDEs, highlight prominent works, and discuss known difficulties and structural limitations. Previous familiarity with machine learning is not assumed.
