Publications
I am currently researching in the Tufts Machine Learning Group and working on submitting a publication in the next few months. This will involve our work on deep multiple instance learning problems with limited labeled data.
Previously, I have worked in computational chemistry, mathmatical modeling, and optimization research labs. My four peer-reviewed publications and two conference posters are shown below.
Synthetic Data Reveals Generalization Gaps in Correlated Multiple Instance Learning Ethan Harvey, Dennis J. Loevlie, and Michael Hughes We evaluate how well current multiple instance learning (MIL) models capture inter-instance correlations by generating synthetic datasets and constructing a Bayes estimator as an optimal upper bound on model performance.
Demystifying the chemical ordering of multimetallic nanoparticles Dennis Johan Loevlie, Brenno Ferreira, Giannis Mpourmpakis A bond-centric model has been shown to capture metal nanoparticle energetics accurately but fails in certain bimetallic combinations, such as Pd and Au. Here we propose an improvement to the bond-centric model that leads to a 71% reduction in the RMSE in the dataset we investigated. We then applied mixed-integer optimization to elucidate the most stable atomic orderings of various multi-metallic nanoparticles.
Single Atom Alloys Segregation in the Presence of Ligands Maya Salem, Dennis J. Loevlie, and Giannis Mpourmpakis The stability of single atom alloys (SAAs) and how it is affected by surface segregation in the presence of ligands is investigated. Using density functional theory (DFT) and machine learning, specifically a neural network model, the study reveals that ligands moderate the thermodynamics of segregation in SAAs. The machine learning model developed in the study accurately predicts the segregation energy (Eseg) in the presence of ligands, aiding in the rapid and efficient screening of SAAs. This research provides insights into the complex physics governing nanoscale interfaces, particularly how ligands influence the segregation behavior in SAAs.
Size-dependent shape distributions of platinum nanoparticles Ruikand Ding, Ingrid M. Padilla Espinosa, Dennis Loevlie, Soodabeh Azadehranjbar, Andrew J. Baker, Giannis Mpourmpakis, Ashlie Martini, and Tevis D. B. Jacobs Experimentally a strong shape dependence was observed as a function of nanoparticle size. Using thermodynamics and Boltzmann statistics, we validate these size-dependent shape distributions, which provides an explanation for this phenomenon.
Resolving electrocatalytic imprecision in atomically precise metal nanoclusters Anantha Venkataraman Nagarajan, Dennis Johan Loevlie, Michael J Cowan, Giannis Mpourmpakis An overview of the most recent developments in electrocatalysis of ligand protected metal nanoclusters. An outlook is provided describing the current and possible future applications of machine learning in the field.
Software Development for HER High-Throughput Experiments Advised by Dr. John Kitchin An overview of software tools developed in Python to increase the efficiency and depth of experimental result analysis. The code developed in Python involved image analysis, search tools, regression and classification, and interactive GUI's. The poster was presented at the Carnegie Mellon University ChemE Masters Student Assosiation Research Forum and was awarded 3rd place.
Mathematical Modeling and Optimization of an Ion Transport Membrane for Oxygen Separation from Air Advised by Dr. Fernando Lima Cryogenic air seperation (CAS) is an extremely energy intensive process. Through mathmatical modeling, economic evaluation, and non-linear, multivariable optimization of an ion transport membrane (ITM) air seperation unit we show that as ITM technology matures it may be a viable alternative to CAS. This research was presented at the 2018 national AIChE conference and won 2nd place in the Computing and Process Control, poster competition.