Multifidelity Machine Learning

Multifidelity machine learning (MFML) leverages data from multiple sources of varying accuracy (fidelities) to build more efficient and robust machine learning models, particularly when high-fidelity data is scarce and expensive to acquire. Current research focuses on developing MFML algorithms, such as optimized MFML and multifidelity Monte Carlo methods, and exploring their application across diverse fields including quantum chemistry, subsurface flow modeling, and solving partial differential equations. This approach significantly reduces the computational cost and improves the accuracy of surrogate models for complex systems, impacting scientific discovery and engineering design by enabling faster and more reliable simulations.

Papers

October 15, 2024

July 24, 2024

Assessing Non-Nested Configurations of Multifidelity Machine Learning for Quantum-Chemical Properties
Vivin Vinod, Peter Zaspel
Chemical Data Quantum Chemistry Molecular Geometry Critical Configuration Multifidelity Machine Learning Quantum Chemical Property

June 20, 2024

QeMFi: A Multifidelity Dataset of Quantum Chemical Properties of Diverse Molecules
Vivin Vinod, Peter Zaspel
Quantum Chemistry Excited State Quantum Chemical Property Multifidelity Machine Learning Multifidelity Estimator Multifidelity Linear Regression

March 13, 2024

Multifidelity linear regression for scientific machine learning from scarce data
Elizabeth Qian, Anirban Chaudhuri, Dayoung Kang, Vignesh Sella
Surrogate Model Scientific Machine Learning Low Quality Scarce Data Multifidelity Machine Learning Multifidelity Estimator Multifidelity Linear Regression

January 15, 2024

Multifidelity domain decomposition-based physics-informed neural networks and operators for time-dependent problems
Alexander Heinlein, Amanda A. Howard, Damien Beecroft, Panos Stinis
Physic Informed Neural Network Feature Operator Domain Decomposition Multifidelity Machine Learning Discretization Invariant

July 23, 2023

Multifidelity Covariance Estimation via Regression on the Manifold of Symmetric Positive Definite Matrices
Aimee Maurais, Terrence Alsup, Benjamin Peherstorfer, Youssef Marzouk
Novel Regression Dimensional Manifold Covariance Estimation Riemannian Classifier Symmetric Positive Definite Matrix Multifidelity Machine Learning Multifidelity Estimator

April 8, 2023

A multifidelity approach to continual learning for physical systems
Amanda Howard, Yucheng Fu, Panos Stinis
Continual LEArning Physic Informed Neural Network Catastrophic Forgetting Continual Learning Physical System Multifidelity Machine Learning

June 10, 2022

Multifidelity Reinforcement Learning with Control Variates
Sami Khairy, Prasanna Balaprakash
Reinforcement Learning High Fidelity Control Variate Multifidelity Machine Learning Multifidelity Estimator

April 23, 2022

Use of Multifidelity Training Data and Transfer Learning for Efficient Construction of Subsurface Flow Surrogate Models
Su Jiang, Louis J. Durlofsky
Transfer Learning Surrogate Model Greater Public Use Data Assimilation Fidelity Model Data Driven Surrogate Deep Learning Based Surrogate Multifidelity Machine Learning Subsurface Model

April 14, 2022

Multifidelity deep neural operators for efficient learning of partial differential equations with application to fast inverse design of nanoscale heat transport
Lu Lu, Raphael Pestourie, Steven G. Johnson, Giuseppe Romano
Partial Differential Equation Efficient Learning Deep Operator Network Inverse Design Thermal Simulation Deep Neural Operator Multifidelity Machine Learning DeepONet Architecture Heat Conduction

Multifidelity Machine Learning

Papers

Investigating Data Hierarchies in Multifidelity Machine Learning for Excitation Energies

Benchmarking Data Efficiency in $Δ$-ML and Multifidelity Models for Quantum Chemistry

Assessing Non-Nested Configurations of Multifidelity Machine Learning for Quantum-Chemical Properties

QeMFi: A Multifidelity Dataset of Quantum Chemical Properties of Diverse Molecules

Multifidelity linear regression for scientific machine learning from scarce data

Multifidelity domain decomposition-based physics-informed neural networks and operators for time-dependent problems

Multifidelity Covariance Estimation via Regression on the Manifold of Symmetric Positive Definite Matrices

A multifidelity approach to continual learning for physical systems

Multifidelity Reinforcement Learning with Control Variates

Use of Multifidelity Training Data and Transfer Learning for Efficient Construction of Subsurface Flow Surrogate Models

Multifidelity deep neural operators for efficient learning of partial differential equations with application to fast inverse design of nanoscale heat transport