Symposium

TYC Symposium

Machine Learning: application to Chemical Reactions

Thursday 25 February 2021
Time: 3-5pm
Venue: Zoom

https://ucl.zoom.us/j/96696928921?pwd=a3htRlVabGJhd3c3UnUzYWxCc3p3Zz09

Meeting ID: 966 9692 8921
Passcode: TYCSymp
Contact: Devis Di Tommaso

Pavlo Dral, Xiamen University

Quantum Chemistry Assisted by Machine Learning

Pavlo will talk about how machine learning (ML) can be used to assist quantum chemical research in a variety of ways and show examples from his research. The latter will include: the Δ-learning and hierarchical ML approach, improving the quantum chemical Hamiltonian with ML, very accurate ML potential energy surfaces, and ML for simulating absorption cross sections and performing nonadiabatic excited-state dynamics.
See more at http://dr-dral.com/research/machine-learning-in-chemistry.

Bio: Pavlo Dral is an Associate Professor at Xiamen University from 2019. He won gold medal in the 36th International Chemistry Olympiad in Germany in 2004. Pavlo Dral received two M.Sc. degrees in 2010: one from University Erlangen-Nuremberg (Germany) in molecular nanoscience and another from the National Technical University of Ukraine “KPI” in chemical technology and engineering of organic compounds. He obtained his PhD in University Erlangen-Nuremberg with Prof. Timothy Clark in 2013 followed by post-doctoral stay with Prof. Walter Thiel in Max-Planck-Institute for Coal Research until 2019. His research area is practical computational chemistry performed with quantum chemical and machine learning methods. See more at http://dr-dral.com.

 

Teodoro Laino, IBM Corporation

Learning the language of organic chemistry: developing artificial intelligence models using existing knowledge

Chemical Synthesis is a fundamental task in organic chemistry.  What if one could learn from all reactions published in articles and patents?  What if this knowledge could be used to build a system that would assist researchers in the synthesis route design process?  The use of data-based models combined with Artificial Intelligence (AI) strategies is emerging as a valuable and robust solution to address these fundamental questions.  Building on the concept of treating chemistry as a language, I will present the use of AI language models to treat forward reaction prediction, retrosynthesis and the prediction of reaction synthesis procedures.  All models are freely accessible world-wide as cloud service: https://rxn.res.ibm.com

Bio: Teodoro Laino is Distinguished Research Staff Member and Manager of Accelerated Discovery at IBM Research Zurich.  He graduated in theoretical chemistry in 2001 from University of Pisa and Scuola Normale Superiore (SNS) di Pisa, and received the doctorate in chemistry in 2006 from SNS.  The focus of his research is on materials simulations for industrial-related problems and on the application of machine learning/artificial intelligence technologies to chemistry and materials science problems.


Luca Ghiringhelli, Fritz Haber Institute

Bridging scales with symbolic inference: the case of heterogeneous catalysis


Heterogeneous catalysis is governed by a multiscale interplay of several processes, e.g., the dynamic re-structuring of the catalyst material at reaction conditions and a network of chemical reactions.  Arguably, modelling the full catalytic progression via first-principles statistical mechanics is impractical, if not impossible.

Here, we propose to use artificial-intelligence, in the form of both predictive and descriptive symbolic inference, in order to bridge the scales and determine the key descriptive parameters (the 'materials genes') encoding the processes that trigger, facilitate, or hinder the catalyst's performance. In practice, symbolic inference learns interpretable models in the form of analytic equations, or inequalities, which predict performance indicators or describe materials that are good candidates for best performance.  We demonstrate the approach by means of two case studies: a) the selective oxidation of propane starting from a dataset of few vanadium-based catalysts.  These materials were carefully synthesized, fully characterized, and tested according to standardized protocols, and b) CO2 conversion on oxide materials, starting from a dataset of ab initio pristine surfaces and CO2 adsorbed on different sites of the oxide surfaces.

This approach highlights the underlying physiochemical processes, and may accelerate catalyst design.

Bio: Dr Luca M. Ghiringhelli leads the group "Big-Data analytics for Materials Science" in the Novel Materials Discovery (NOMAD) Laboratory at Fritz Haber Institute of the Max Planck Society, in Berlin.  Formerly, he has led the group "Ab initio statistical mechanics of cluster catalysis and corrosion" in the theory group at the same institute.  His background is in computational statistical mechanics and electronic structure methods, applied to the evaluation of thermodynamic and kinetic properties of bulk materials, surfaces, and nano-clusters.  Within the NOMAD Lab, he leads the development and application of methods based on compressed sensing, symbolic regression, subgroup discovery, and deep learning to the modelling of big data in materials science.  His focus is on methods that yield interpretable models and can cope with "small data" for training.  He also co-led the development of the hierarchical and extensible metadata infrastructure for the NOMAD Lab.

Since January 2018, he is co-leading the Psi-k working group on "High-throughput screening and data analytics".

6yBGvF_web.jpg

Follow @tyc_london for updates from the Thomas Young Centre.