AI for Big Data Problems in Imaging, Modeling and Synthesis: AI-accelerated Materials Discovery and Synthesis
Program Organizers: Mathew Cherukara, Argonne National Laboratory; Badri Narayanan, University of Louisville; Subramanian Sankaranarayanan, University of Illinois (Chicago)
Wednesday 8:00 AM
November 4, 2020
Room: Virtual Meeting Room 41
Location: MS&T Virtual
Session Chair: Subramanian Sankaranarayanan, University of Illinois (Chicago); Badri Narayanan, University of Louisville; Mathew Cherukara, Argonne National Laboratory
8:00 AM
Enabling Data-driven Discovery of Chemistry-function Relationships via Automated Packing Motif Labeling: Donald Loveland1; Phan Nguyen1; Anna Hiszpanski1; T. Yong-Jin Han1; 1Lawrence Livermore National Laboratory
Models that predict bulk properties of molecular materials from chemical structures alone and apriori to synthesis are desired to accelerate materials development. While chemistry-based machine learning has advanced these efforts, insufficient data remains a challenge. The packing motif, a visually distinctive pattern in how aromatic molecules are oriented relative to one another in a crystal structure, is an example characteristic that influences many bulk properties but lacks large datasets for machine learning. Efforts to create such datasets using automated labeling tools by geometric descriptors have been stymied due to the difficulty of selecting the appropriate crystal orientation for the packing motif to be evident. We developed a procedure to identify the appropriate crystallographic plane for analysis, improving packing motif label accuracy by up to 25% compared to previous methods. With the ability to construct large motif datasets, we investigate intermolecular interactions that correlate with specific motifs helping guide synthesis efforts.
8:20 AM Invited
Directing Matter In-situ via Deep Learning: Bobby Sumpter1; 1Oak Ridge National Laboratory
Recent advances in computational algorithms and computer capacities that are orders of magnitude larger and faster, have enabled extreme-scale simulations and deep data analytics of materials properties and chemical processes. This powerful confluence of capabilities and the information bound in large volumes of high-quality data offers exciting new opportunities for accelerating design and discovery of materials. In this talk I will discuss how we are now probing in-situ, chemical reactions and materials transformations as a modality for direct feedback to an experiment in order to precisely impart directed energy (electrons, ions) that manipulates a material at the nanoscale. This approach is enabled via the dual capability of high-resolution experimental imaging and focused energy in-situ, providing data rates, quality and volumes that allow a deep learning framework to accurately identify materials structures and dynamics across broad length and time scales.
9:00 AM
Inverse Design of Porous Structures by Deep Learning and TPU-based Computing: Yuhai Li1; Yuhan Liu1; Mathieu Bauchy1; 1University of California, Los Angeles
Although simulations offer a convenient pathway to predict the properties of a given structure, “inverse design” optimizations (i.e., predicting which structure exhibits the most desirable properties) are notoriously challenging problems due to the vastness of the design space. Here, we present a deep learning framework that greatly accelerates the discovery of promising structures featuring optimal mechanical properties. Our approach relies on a convolutional neural network (CNN) model (trained from hight-hroughput peridynamic simulations) that successfully maps a structure to its associated stress-strain curve upon tensile fracture. The CNN predictor is then used to train an inverse CNN generator model enabling the prediction of optimal structures. As a key enabler of this approach, we adopt Tensor Processing Unit (TPU) computing, which offers unprecedented performance in training large, complex neural networks. We suggest that TPU-based deep learning offers a new pathway to accelerate the discovery of novel materials with exotic properties and functionalities.
9:20 AM Invited
Polymer Informatics—Current Status and Critical Next Steps: Lihua Chen1; Rampi Ramprasad1; 1Georgia Institute of Technology
The Materials Genome Initiative (MGI) has heralded a sea change in the philosophy of materials design. In an increasing number of applications, the successful deployment of novel materials has benefited from the use of computational, experimental and informatics methodologies. Here, we describe the role played by computational and experimental data generation and capture, polymer fingerprinting, machine-learning based property prediction models, and algorithms for designing polymers meeting target property requirements. These efforts have culminated in the creation of an online Polymer Informatics platform (https://www.polymergenome.org) to guide ongoing and future polymer discovery and design. Challenges that remain will be examined, and systematic steps that may be taken to extend the applicability of such informatics efforts to a wide range of technological domains will be discussed. These include strategies to deal with the data bottleneck, new methods to represent polymer morphology and processing conditions, and the applicability of emerging algorithms for design.