Late News—Hot Topics Abstracts were due September 3.
Authors will be notified by mid-September.
Symposium CH04-Accelerating Materials Characterization, Modeling, and Discovery by Physics-Informed Machine Learning
Machine learning methods are making rapid inroads into all fields of science, driven by data volumes, computational resources, and utility of the methods at finding correlations in high dimensional spaces. Both methods (algorithms) and infrastructure (combinatorial and high-throughput experimentation, high-performance computing, databases, data-management systems, workflows, repositories) are being developed to both create and tackle the large increase in data volume, facilitating the leap from single lab-based experiment-computation-output models to one where researchers can utilize available public materials data infrastructure, leading to new materials discoveries, enhanced predictive capabilities, and accelerated scientific understanding.
This symposium will bring together the latest advances in the development and application of machine learning and related data analytics methodologies to enhance the characterization of materials, extract relevant features for improving theory-experiment comparisons, and assist in solving inverse problems relevant to structural and functional characterization.Further, it will encompass robust uncertainty quantification approaches, such as the use of Bayesian methods for challenging characterization and prediction tasks and enabling automated experimentation, as well as discuss methods that extend beyond traditional correlative machine learning methods towards causal inference to better understand drivers of materials’ behaviors.
Topics will include:
- High-throughput materials synthesis for the generation of large and consistent datasets
- High-throughput characterization and computations for materials discovery
- New techniques and methods enabled by machine learning approaches for probing and characterizing the structural, chemical and/or electronic nature of materials
- Bridging computation and experimental data via machine learning and statistical methods, including solving inverse problems, feature extraction and selection, and materials design
- Materials data infrastructure – databases, data-management systems, workflows and best practices for 21st century materials science
- Causal inference and Bayesian models for incorporating prior information, model selection, and uncertainty quantification
- Reinforcement learning and Gaussian process methods for automated experimentation, materials design, synthesis and characterization
- A tutorial complementing this symposium is tentatively planned.
(University of Toronto, Canada)
(Rutherford Appleton Laboratory, United Kingdom)
(Duke University, USA)
(Humboldt-Universität zu Berlin, Germany)
(Argonne National Laboratory, USA)
(Aalto University, Finland)
(California Institute of Technology, USA)
(Carnegie Mellon University, USA)
(Oak Ridge National Laboratory, USA)
(Citrine Informatics, USA)
(Massachusetts Institute of Technology, USA)
(University of California, Berkeley, USA)
(Georgia Institute of Technology, USA)
(Fritz Haber Institute of the Max Planck Society, Germany)
(Commissariat à l’énergie atomique et aux énergies alternatives, France)
Helmholtz-Zentrum Berlin für Materialien und Energie
Institute Functional Oxides for Energy-Efficient Information Technology
Argonne National Laboratory
Center for Nanoscale Materials
National Institute of Standards and Technology
Thermodynamics and Kinetics Group
Oak Ridge National Laboratory
Center for Nanophase Materials Sciences