Accelerate Materials Discovery and Phenomena
Recent advances in machine learning have created a revolution in all areas of science and engineering. For example, advanced data analytics tools based on computer vision methods such as object detection and image segmentation are able to characterize millions of experimental images generated by instruments like scanning electron microscope. On the other hand, deep learning (DL) models designed using active learning, reinforcement learning and generative models such as variational autoencoder are able to autonomously navigate complex materials energy landscapes to discover new molecules with desired properties, predict reaction pathways and optimal conditions for chemical reactions with little to no human supervision. Further, machine learning methods combined with atomistic modeling and accelerated dynamics has enabled high throughput screening of materials and reach sufficiently long-time scale material simulation to study rare events. Availability of exascale computers due to arrive soon will make it easier to model hard and soft materials and biological systems with deep learning in conjunction with molecular dynamics (MD) simulations. Billion-to-trillion atom MD simulations with DL trained on ab initio quantum mechanical simulations can reliably describe charge transfer, bond breaking/bond formation, and chemical reactions in materials under normal and extreme operating conditions.
Topics covered include, but are not limited to:
- Machine learning for energy landscapes and force field development
- Materials design using variational auto encoder and generative adversarial network
- Polymer design using machine learning
- Reinforcement learning for synthesis of quantum materials
- Neural network quantum molecular dynamics
- Accelerated dynamics using reinforcement learning
Guest Editors
Priya Vashishta, University of Southern California
Rajiv K Kalia, University of Southern California
Aiichiro Nakano, University of Southern California
Roberto Car, Princeton University
Nicola Marzari, EPFL
APL Editor
Vincenzo Lordi, Lawrence Livermore National Laboratory