Patching Analytic Solutions with AI: The GlueNN Framework

by Dr Jiheon Lee (KAIST)

Thursday 30 Apr 2026, 16:00 → 17:30 Asia/Seoul

Room 1423 (KIAS)

Title: Patching Analytic Solutions with AI: The GlueNN Framework

Speaker: Jiheon Lee

Abstract: 

Many problems in physics involve multiple dynamical regimes, where the objective is not only to obtain accurate solutions but also to understand how different physical behaviors emerge and connect across scales. Conventional numerical solvers and many machine-learning approaches, including standard Physics-Informed Neural Networks (PINNs), typically learn the full solution directly and often provide limited interpretability.

In this talk, I will introduce GlueNN, a physics-informed framework that constructs global solutions by combining local analytic forms and learning the scale-dependent coefficients that interpolate between them. GlueNN yields smooth transitions between regimes without ad hoc matching procedures, while preserving interpretability and enabling direct extraction of physically meaningful quantities.