Symposium DS02-Integrating Machine Learning with Simulations for Accelerated Materials Modeling

The advent of high-speed computation has significantly accelerated the materials modeling and simulation paradigm in the past few decades. These simulations cover wide length-and time-scales using ab-initio first principle, atomistic, mesoscale, and continuum simulations. Despite the economical nature of these simulations in comparison to experiments, they still suffer from the major deficiencies, namely, limitations on the systems size, simulation time, accuracy of simulations, transferability of a simulation to different scenarios, to name a few. For example, while first principle simulations can provide accurate predictions on the material response to the electronic level, they are limited to a few hundreds of atoms. Recently, machine learning has shown promising means to address some of these challenges successfully. Some of these developments include machine-learned interatomic potentials, physics-informed neural networks, convolutional neural network based microstructure modeling, and graph neural networks for structure–property correlations. This symposium will highlight the latest development in machine learning for materials simulations with specific focus in three major areas: (i) supporting and accelerating simulations using machine learning (for example, machine learned potentials), (ii) interpreting and decoding simulations and high-throughput data using machine learning, (iii) replacing traditional differential equation-based simulations with machine-learned simulations.

Topics will include:

Development of machine learned interatomic potentials
Physics-informed machine learning models for materials simulation
Graph neural network; Transfer learning; Using natural language processing, for materials modeling
Development of realistic material models using image processing
Topology optimization and development of tailored microstructure using machine learning
Reduced order machine learning models for atomistic simulations
Machine learning for coarse-grained and continuum simulations
Use of synthetic (or simulated) data to train machine learning models
Multi-fidelity/data-fusion approaches to combine synthetic and experimental datasets
Automated identification of collective variables for metadynamics approaches
Use of hardware dedicated to deep learning (e.g., TPUs) to accelerate simulations
End-to-end differentiable simulations and meta-optimization using machine learning

Invited Speakers:

Keith Brown (Boston University, USA)
Josephine Carstensen (Massachusetts Institute of Technology, USA)
Maria K. Chan (Argonne National Laboratory, USA)
Shu-Wei Chang (National Taiwan University, Taiwan)
Gowoon Cheon (Google, USA)
Chiara Daraio (California Institute of Technology, USA)
Krishnakumar Garikipati (University of Michigan, USA)
Jonathan Godwin (Deep Mind, United Kingdom)
Jeffrey Grossman (Massachusetts Institute of Technology, USA)
Elizabeth Holm (Carnegie Mellon University, USA)
Karsten Jacobsen (Technical University of Denmark, Denmark)
Sinan Keten (Northwestern University, USA)
George Marchant (University of Warwick, United Kingdom)
Arif Masud (University of Illinois at Urbana-Champaign, USA)
Aiichiro Nakano (University of Southern California, USA)
Harold Park (Boston University, USA)
Ghanshyam Pilania (Los Alamos National Laboratory, USA)
Seunghwa Ryu (Korea Advanced Institute of Science and Technology, Republic of Korea)
Subramanian Sankaranarayanan (Argonne National Laboratory, USA)
Sam Schoenholz (Google Research, USA)

Symposium Organizers

N. M. Anoop Krishnan

Indian Institute of Technology Delhi

Civil Engineering

India

Mathieu Bauchy

University of California, Los Angeles

Civil and Environmental Engineering

USA

Ekin Dogus Cubuk

Google Brain

USA

Grace Gu

University of California, Berkeley

Mechanical Engineering

USA

[email protected]

Topics

artificial intelligence machine learning materials genome modeling simulation