Scientific Machine Learning Lab
Scientific Machine Learning (SciML) is an emerging area that brings together the fields of Machine Learning and Scientific Computation. SciML introduces scientific model constraints in Machine Learning algorithms, allowing prediction of future performance of complex multiscale, multiphysics systems using sparse, low-fidelity, and heterogeneous data. Unlike traditional black-box Machine Learning methods, SciML aims to deliver interpretable models, leading to improved verification and validation in mission-critical applications
The Scientific Machine Learning Lab (SciML Lab) was created to support and grow a community of researchers across Texas A&M involved in the development of Scientific Machine Learning algorithmic, computational, and applied components. The SciML Lab aims at bringing together a multidisciplinary team of scientists, mathematicians, physicists and engineers that cut across several Colleges and Departments at Texas A&M, who share a unified goal of building physics-aware machine learning architectures from the underlying physical phenomena. The Lab will will push the boundaries of each individual’s discipline and will serve as a catalyst for accelerating education in SciML, through short hands-on courses, seminars, workshops, and case studies. SciML Lab was established to pilot the Thematic Data Science Labs program of the Texas A&M Institute of Data Science (TAMIDS).
This course introduces the foundations of Scientific Machine Learning (SciML), a rapidly developing area that brings together the fields of Machine Learning and Scientific Computation. After an introduction to the field, we review classical ODE and PDE discretization methods. Next, we discuss automatic differentiation and introduce the main physics-informed methods in SciML, namely, PDE-constrained neural network regression and PDE-constrained Gaussian Process regression. We then present data-driven SciML methods, including operator learning, foundation models, and data-driven reduced-order models. We include throughout the application of SciML methods in forward prediction, inverse modeling, and uncertainty quantification.
Research Projects
Our goals in this project concern new architectures and new algorithms for physics-informed neural networks, neural operators, and Bayesian models.
This project explores new approaches to physics-based and data-driven models and combinations thereof that effectively embeds physical constraints, conservation laws and constitutive relations with data for robust upscaling, sensible time-stepping, accurate proxy models and efficient optimization schemes.
This project will lead to new approaches to Type Ia supernovae (SN~Ia) cosmology that can quantitatively incorporate the latest in observational and theoretical developments using SciML tools.
This project is a recent collaboration with the Institute of Fusion Sciences at UT-Austin, Virginia Tech, and VTT (Finland). The goal is to develop deep learning surrogates for fast simulation of plasma turbulence in fusion devices.
The goal of this project is to develop novel scientific machine learning (SciML) algorithms and to apply them to highly coupled multi-physics thermodynamically consistent phase field models for microstructure evolution, in order to deliver a set of efficient and accurate physics-aware forward and inverse models of processing-microstructure relationships in nanostructured composite thermoelectric materials, together with reliable uncertainty quantification metrics to support decision making in the design of these materials.
Resources
Contact
Collaboration requests should be made to individual SciML Lab researchers.
Ulisses Braga-Neto
Organizational questions concerning SciML Lab
ContactLevi McClenny
Website update requests for research, events, or announcements
Contact