Susanta Ghosh
Michigan Technological University, MI·Engineering·Early-career
Where electrons sit inside a material determines nearly everything about it: its strength, conductivity and chemistry. Calculating that electron distribution from first principles is accurate but so costly that realistic system sizes are out of reach.
- Fields
Engineering
Materials Science
Computer Science
- Topics
electronic structure
electron density prediction
machine learning
- Grants
- 1 known grant, through Aug 2030
What they work on
- Develop a Bayesian convolutional neural network that predicts electron density, the map of where electrons reside in a material, while reporting its own uncertainty.
- Make the model equivariant to rotation and translation, so that reorienting or shifting a set of atoms transforms the predicted electron density consistently, and make it physics-informed through volumetric inputs and outputs.
- Create a technique for sampling thermo-mechanical variables during training-data generation, plus data pruning schemes, to improve generalization and cut the cost of data generation and training.
- Demonstrate accuracy and scalability across metals, alloys with and without defects, and molecules, then extend the model to a wide range of transition metals and their alloys.
- Use the model's uncertainty estimates within a Bayesian Optimization framework to guide the design of high-entropy alloys, materials mixing several metals in near-equal proportions.
More about the research
Machine learning can stand in for those calculations, but existing models typically fall short on at least one requirement: honest estimates of their own uncertainty, respect for the symmetries of physics, grounding in physical laws, generalization to new compositions, accuracy, speed, or the ability to scale up. This project attempts to satisfy all of these within one framework for predicting electron density, the map of where electrons are likely to be found.
Why it matters
Designing new metals, alloys and devices is slowed by the cost of electronic-structure simulation at useful sizes. A fast, symmetry-respecting predictor of electron density with trustworthy uncertainty estimates could let engineers screen candidate high-entropy alloys computationally before making them. The same combination of uncertainty quantification and rotational-translational symmetry is relevant to biomedical imaging and continuum physics problems.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Qi An
Iowa State University, IA·Materials Science
Build a closed-loop AI platform that compares Bayesian optimization (choosing the most informative next test) with deep reinforcement learning (trial-and-reward learning) for designing the high-entropy carbide (Ti-Zr-Hf-Nb-Ta)C, using machine-learning force-field atomistic simulations.
Qi Yu
University of California-San Diego, CA·Computer Science·Early-career
Develop deep learning models that forecast spatiotemporal dynamics while obeying physical laws, with the goal of simulating ocean currents, traffic and epidemic spread faster than numerical solvers.
Jaime Marian
University of California-Los Angeles, CA·Materials Science
Determine how 'necklace' microstructures nucleate at grain boundaries during dynamic recrystallization in body-centered-cubic refractory multi-principal element alloys, using mesoscale modeling.
The school
Michigan Technological University: 9 labs on PI Pages, $107M a year on research, 86 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.