Muammar El Khatib
Muammar El Khatib is a postdoctoral scholar in the Computational Chemistry, Materials and Climate group at Lawrence Berkeley National Laboratory (LBL).
He is a chemist by training from the University of Zulia in Venezuela and started his graduate studies with a European Master in Theoretical Chemistry and Computational modeling of the Erasmus Mundus Program. His Ph.D. in theoretical chemical physics was about the characterization of metallic and insulating properties of low-dimensional systems using the theory of the insulating state of Walter Kohn applied with wave function theory.
Prior to LBL, he was a postdoctoral research associate at Brown University where he worked in the acceleration of atomistic simulations with machine learning models in the group of Prof. Andrew A. Peterson in the Catalyst Design Laboratory. In this appointment, he acquired experience with neural network and kernel ridge regression models to mimic quantum mechanics simulations using interatomic machine learning potentials.
He has published 10 papers, given presentations at international conferences, and developed the ML4Chem machine learning package, a module for the Molpro quantum chemistry package, and the atomistic machine-learning package (Amp). Additionally, he has participated in the free software community and is a Debian Linux developer.
At LBL, he is working towards the development of machine learning approaches, algorithms and data sets to solve chemical sciences problems.
Muammar El Khatib, Wibe A de Jong, "ML4Chem: A Machine Learning Package for Chemistry and Materials Science", March 9, 2020, doi: 10.26434/chemrxiv.11952516.v1
Oriana Brea, Muammar El Khatib, Celestino Angeli, Gian Luigi Bendazzoli, Stefano Evangelisti, Thierry Leininger, "The Spin-Partitioned Total-Position Spread: an application to diatomic molecules", J. Phys. Chem. A, 120, 5230 (2016), March 25, 2016,
Muammar El Khatib, Oriana Brea, Edoardo Fertitta, Stefano Evangelisti, Thierry Leininger, Gian Lugi Bendazzoli, "The total position- spread tensor: spin partition", J. Chem. Phys. 142, 094113 (2015), March 5, 2015,
Muammar El Khatib, Gian Luigi Bendazzoli, Stefano Evangelisti, Wissam Helal, Thierry Leininger, Lorenzo Tenti, Celestino Angeli, "Beryllium- Dimer: a Bond based on non-Dynamical Correlation", J. Phys. Chem. A, 6664 (2014), May 27, 2014,
Muammar El Khatib, Wibe De Jong, Feature Extraction Using Semi-Supervised Deep Learning., APS March 2020, March 5, 2020,
Features are defined as measurable properties that characterize observed phenomena and represent a key part of machine learning (ML) algorithms. In materials sciences, ML has successfully accelerated atomistic simulations using man-engineered features for tasks such as energy or atomic forces predictions. These features fulfill physics constraints such as rotational and translational invariance, uniqueness and, locality (the sum of local contributions reconstructs a global quantity). However, these ML models are known to perform poorly when operating out of the training set regime because features are not representative of the underlying structure of the data. This could be improved if features are extracted with advanced hybrid architectures e.g. a variational autoencoder that is trained with physics constraints introduced with an external task and a loss function. We will explore how the use of semi-supervised learning techniques can be a powerful tool for the extraction of features for atomistic simulations. All results shown herein can be reproduced with ML4Chem: a free software package for machine learning in chemistry and materials sciences.