alee

Dr. Alpha Lee – A3MD Distinguished Seminar Series

Dr. Alpha Lee delivered an A3MD Distinguished Seminar entitled: “Accelerating the materials design-make-test cycle with machine learning and coarse graining

Unlike molecular chemistry, where valance-bond theory provides a robust framework to represent molecules as a graph, representing inorganic materials is significantly more intricate. On the one hand, representing a material using composition alone removes the possibility of identifying polymorphs. On the other hand, a representation based on distances between atoms is challenging to deploy in a computational high throughput screening workflow for novel materials because it is a priori unclear whether a particular composition/structure is thermodynamically stable. In my talk, I will discuss our journey in applying concepts in coarse graining to devise machine learning models for materials properties prediction and materials synthesis prediction. I will first discuss the lowest level of coarse graining – stoichiometry – and discuss a representation learning framework that predicts materials properties using stoichiometry as input. I will then show how this framework can be extended to predict the outcomes of materials synthesis, whilst also providing an interpretable “reaction similarity” metric that enables rapid search for literature precedents in materials synthesis. Finally, I will move up the coarse graining ladder and discuss how we can incorporate structure, albeit in a coarse-grained way, using the concept of Wyckoff representation. This approach allows us to effectively amortise crystal structure searching, turning an infinite search space into a combinatorially enumerable search problem.

0-Student-John-Gregoire_bobpaz.com0137-WEB.original

Professor John Gregoire – A3MD Distinguished Seminar Series

Professor John Gregoire delivered an A3MD Distinguished Seminar entitled: “Accelerating discovery of solar fuels materials with high throughput experiments and artificial intelligence

The large data flux of high throughput experiments naturally presents opportunities for data science. Addressing the most pressing materials discovery challenges with these techniques requires design of experiments and algorithms in the context of the materials physics and chemistry that are pertinent to the target technology. The corresponding quest for materials that harvest solar energy to generate chemical fuels has led to research problems that challenge the state of the art in artificial intelligence. The ensuing efforts to generate new algorithms and modalities of research include 1) the development of deep reasoning networks that incorporate physics rules in machine learning, which is critical for automating crystal structure phase mapping; 2) the incorporation of hierarchical correlation learning in multi-property prediction, which conditions models to facilitate prediction in never-before-seen composition spaces; and 3) probabilistic modeling of data in high order composition spaces, which identifies materials most likely to exhibit exceptional properties for any target technology. While these efforts arose from specific research challenges, the methods and concepts are intended to be generally applicable for accelerating scientific discovery via adaption and improvement by the research community.

Laura-00552

Professor Laura Gagliardi – A3MD Distinguished Seminar Series

Professor Laura Gagliardi delivered an A3MD Distinguished Seminar entitled: “Theoretical and Computational Challenges in Modeling MOF-Based Catalysis and Water Harvesting

Metal-organic frameworks (MOFs) are versatile platforms with tunable properties ranging from high selectivity in gas separations, to catalytic activity for complex reactions, to unique magnetic properties. In collaboration with experimentalists, we try to understand the activity of MOF-based catalysts for reactions related to natural gas conversion, e.g., catalytic oligomerization of abundant C1, C2, and C3 hydrocarbons to longer congeners, or their selective oxidation to alcohols or other fuel molecules. Modeling these species poses enormous challenges from a theoretical and computational perspective. I will describe our latest results in modeling light-alkane hydroxylation over Fe-based MOFs. I will also describe our combined computational and data-science approach to explore MOFs quantum-chemical properties.

Headshot_YSJ3

Professor Yousung Jung – A3MD Distinguished Seminar Series

Professor Yousung Jung delivered an A3MD Distinguished Seminar entitled: “AI-Accelerated Materials Structure-Property-Synthesizability Prediction

The constant demand for new functional materials calls for efficient strategies to accelerate the materials design and discovery. In addressing this challenge, materials informatics deals with the use of data, informatics, and machine learning (complementary to expert’s intuitions) to establish the materials structure-property relationships and to make a new functional discovery in a rate that is significantly accelerated. In forward mapping, one aims to predict the materials properties using their structures as input, encoded in various ways such as simple attributes of constituent atoms, compositions, structures in graph forms, and etc, while in inverse mapping, one defines the desired property first and attempts to find the materials with such property. In this talk, I will mainly focus on the method of machine learning for fast and reliable evaluation of materials properties with a particular emphasis on catalysis applications, and also describe a method to predict the synthesizability of inorganic crystals based on structural encoding.

Headshot

Professor Geoffroy Hautier – A3MD Distinguished Seminar Series

Professor Geoffroy Hautier delivered an A3MD Distinguished Seminar entitled: “Finding the needle in the haystack: accelerated identification of materials with exceptional opto-electronic properties

Essential materials properties can now be assessed through ab initio methods. When coupled with the exponential rise in computational power, this predictive power provides an opportunity for large-scale computational searches for new materials. We can now screen thousands of materials by their computed properties even before starting any experimental work. This computational paradigm allows experimentalists to focus on the most promising candidates and enable researchers to efficiently and rapidly explores new chemical spaces. In this talk, I will present how this approach has been used to identify unexpected materials with exceptional opto-electronic properties. I will especially focus on some of our latest work in several fields from new p-type transparent conducting materials to electrides. I will outline the challenges and opportunities in the field of high-throughput materials discovery and discuss the growing use of machine learning and materials databases such as the Materials Project.

Jensen_crop

Professor Jan Jensen – A3MD Distinguished Seminar Series

Professor Jan Jensen of U. Copenhagen delivered an A3MD Distinguished Seminar entitled: “Chemical Space Exploration”

I’ll talk about how we use quantum chemistry, genetic algorithms, and machine learning to search chemical space for molecules with specific properties. Examples include molecules that absorb light at specific wavelengths, bind to protein targets, or catalyse reactions. I’ll also discuss how we help ensure that the molecules are synthetically accessible.

Persson_crop

Professor Kristin Persson – A3MD Distinguished Seminar Series

Professor Kristin Persson of Lawrence Berkeley National Lab delivered an A3MD Distinguished Seminar entitled: “The Era of Data-driven Materials Innovation and Design

Fueled by our abilities to compute materials properties and characteristics orders of magnitude faster than they can be measured and recent advancements in harnessing literature data, we are entering the era of the fourth paradigm of science: data-driven materials design. The Materials Project (www.materialsproject.org) uses supercomputing together with state-of-the-art quantum mechanical theory to compute the properties of all known inorganic materials and beyond, design novel materials and offer the data for free to the community together with online analysis and design algorithms. The current release contains data derived from quantum mechanical calculations for over 100,000 materials and millions of properties. The resource supports a growing community of data-rich materials research, currently comprising over 170,000 registered users and between 2-5 million data records served each day through the API. The software infrastructure enables thousands of calculations per week – enabling screening and predictions – for both novel solid as well as molecular species with target properties.  However, truly accelerating materials innovation also requires rapid synthesis, testing and feedback. The ability to devise data-driven methodologies to guide synthesis efforts is needed as well as rapid interrogation and recowdwrding of results – including ‘non-successful’ ones. In this talk, I will highlight some of our ongoing work, including efficient harnessing of community data together with our own computational data enabling iteration between ideas, new materials development, synthesis and characterization as enabled by new algorithmic tools and data-driven approaches.

Elsa-Olivetti-Laptop-MIT

Professor Elsa Olivetti – A3MD Distinguished Seminar Series

Professor Elsa Olivetti of MIT delivered an A3MD Distinguished Seminar entitled: “Bridging the Gap Between Literature Data Extraction and Domain Specific Materials Informatics

Data has become a fundamental ingredient for accelerating and optimizing materials design and synthesis. Advances in applying natural language processing (NLP) to material science text has greatly increased the size and acquisition speed of materials science data from the published literature. This presentation will describe work to extract information from peer reviewed academic literature across a range of materials. Applying NLP pipelines to these types of materials science systems can be challenging due to the general schema and the noisiness of automatically extraction data. I will present data engineering techniques and discuss an optimal balance between automatic and manual data extraction.

Brabec2

Professor Christoph Brabec delivers Dec 2020 A3MD Distinguished Seminar Series

Professor Christoph Brabec of FAU Erlangen-Nürnberg delivered the Dec 2020 A3MD Distinguished Seminar Series talk entitled: “AMANDA – Line 1: Can AI guided high throughput device engineering resolve long time challenges in solution processed photovoltaics?” 

Evaluating the potential of organic photovoltaics materials and devices for industrial viability is a multi-dimensional large parameter space exploration. Manual experimentation is extremely limited in throughput and reproducibility. Automated platforms for fabricating and characterizing complete functional devices can accelerate experimentation speed within tight processing parameter variations. Here we demonstrate a multi-target evaluation of organic and perovskite photovoltaic materials in full device level with the automated platform AMANDA Line 1 combined with Gaussian progress regression-based data evaluation. Around 100 processing variations are screened within 70 hours which yield a reliable evaluation output in terms of efficiency and photostability. The unprecedented quality of the data coming from the AMANDA platform allow building correlation models by AI methods like Gaussian Parameter Regression (GPR). Already several hundred samples allowed to research for hidden parameter correlations revealing structure – property correlations. One surprising correlation established a direct link between the absorption spectrum of a semiconductor composite and the performance and lifetime of a photovoltaic device. Such correlations have been previously searched for by highly complex experiments, including microstructure investigations on the synchrotron, but haven´t passed the level of qualitative predictions. With AMANDA we have been able to build a quantitative correlation based on simple absorption spectroscopy. The implications of this research concept on the long time challenges in emerging photovoltaics will be discussed in the outlook of the talk.

JlbXKuMO

Dr. Shijing Sun – A3MD Distinguished Seminar Series

Dr. Shijing Sun delivered the inaugural A3MD Distinguished Seminar entitled: “Data-Driven Discovery in the Search for Next-Generation Solar Cell Materials

There is a need for rapid technological development of new sustainable energy technologies to meet climate targets. However, functional clean energy materials comprise a large, high-dimensional space that spans chemical composition and structure, fabrication conditions, and device performance. To tackle this space, we employ theory, data-driven methods, and high-throughput experimentation. In this talk, I discuss efforts using these methods to search for new stable metal halide perovskite photoabsorbers, the design of protective capping layers, and the discovery of new non-toxic perovskites.