Scientific machine learning (SciML) is an emerging topic that has lead to several interesting innovations in recent years, yet the application of machine learning techniques to the study of scientific problems bears many challenges. Among others, off-the-shelf machine learning models often lack interpretability and explainability. To address this challenge, SciML aims to break black-box models into simpler building blocks that can then be studied using, for instance, traditional techniques from dynamical systems and control. Further, integrating prior knowledge about the problem under consideration helps to improve the explainability of predictions and to reduce uncertainties. The Pittsburgh-Berkeley-Stockholm joint e-Seminar on SciML meets biweekly to discuss current methods at the intersection of machine learning and dynamical systems. This seminar aims to explore challenges and solutions for more robust and interpretable models that yield explainable predictions. The seminar is hosted by

Please, sign up for our SciML Google Group in order to receive updates and announcements!

Our meetings are held on Zoom: https://pitt.zoom.us/j/98228841116

Fall semester 2022: 9 am (PST) / 12pm (EST) / 6pm (CET)

  • Friday, September 9, 2022: Lu Lu (University of Pennsylvania)
    • Title: Learning operators using deep neural networks for multiphysics, multiscale, & multifidelity problems (video)
  • Friday, September 23, 2022: Omer San (Oklahoma State University)
    • Title: Hybrid physics-data modeling in fluid dynamics (video)
  • Friday, October 7, 2022: Jan Drgona (PNNL)
    • Title: Differentiable Programming for Modeling and Control of Dynamical Systems (video)
  • Friday, October 21, 2022: Themistoklis Sapsis (MIT)
    • Title: Likelihood-weighted active learning with application to Bayesian optimization, uncertainty quantification, and decision making in high dimensions
  • Friday, November 4, 2022: Youngsoo Choi (LLNL)
    • Title: Interpretable and physics-constrained data-driven methods for physical simulations
  • Friday, November 18, 2022: Lars Ruthotto (Emory University)
    • Title: Neural Network Approaches for High-Dimensional Optimal Control
  • Friday, December 16, 2022: Shams Basir (University of Pittsburgh)
    • Title: Integrating Noisy Data, High Fidelity Data and Governing Laws to Extract Physical Solutions using Artificial Neural Networks

Spring semester 2022: 9 am (PST) / 12pm (EST) / 6pm (CET)

  • Tuesday, January 25, 2022: Weinan E (Peking University)
    • Title: A Mathematical Perspective of Machine Learning (video)
  • Friday, February 4, 2022: Alex Townsend (Cornell University)
    • Title: Learning Green’s functions associated with elliptic PDEs (video)
  • Friday, February 18, 2022: Cyril Zhang (Microsoft Research, NYC)
    • Title: Understanding Neural Net Training Dynamics in Tractable Slices (video)
  • Friday, March 4, 2022: Bao Wang (University of Utah)
    • Title: How Differential Equations and Random Graph Insights BenefitDeep Learning
  • Friday, March 18, 2022: Houman Owhadi (Caltech)
    • Title: Solving/learning nonlinear PDEs and completing computational graphs with GPs (video)
  • Friday, April 1, 2022: Rose Yu (UC San Diego)
    • Title: Exploiting Symmetry in Deep Dynamics Models For Improved Generalization (video)
  • Friday, April 15, 2022: William Gilpin (UT Austin)
    • Title: Comparing statistical forecasting models across a space of chaotic systems (video)
  • Friday, April 29, 2022: Lorenzo Livi (University of Manitoba)
    • Title: Training and Dynamics of Recurrent Neural Networks (video)
  • Friday, May 27, 2022: Mohammad Farazmand (NC State University)
    • Title: Data-driven prediction of dynamical systems from partial observations (video)