Corey Adams, Ph.D. – Corey Adams

Interests: Deep Learning, High Performance Computing, AI for Physics

Background: Accomplished computational scientist with a strong foundation in fundamental science and a track record of applying cutting-edge AI methodologies to diverse projects. A decade of experience with performance optimization for AI and HPC, from single device to supercomputer scale. Proficient in integrating AI solutions into complex computing environments, pushing the boundaries of high-performance computing and contributing to the intersection of AI, machine learning, and computational science.

Skills

Deep Learning: Computer Vision (CNN, GNN, Classification, Detection, Segmentation, UNet, YOLO), Sparse Neural Networks, Generative, Unsupervised and Self-Supervised Learning (GANs, AEs, SimCLR), Surrogate Models, Neural Operators, Distributed Learning and Inference, Scaling, Graph Neural Networks, Machine Learning, Pattern Recognition.

Languages and Frameworks: PyTorch (including FSDP, DTensor), Python, C++, C, CUDA-Python, TensorFlow, Keras, JAX.

Distributed & HPC: MPI, NCCL/RCCL/oneCCL, Slurm/PBS, HDF5/Zarr, Docker/Enroot.

Profiling and Performance: Nsight Systems, VTune, PyTorch Profiler, cProfile/line_profiler (python), mixed precision, CUDA graphs, Triton, NVIDIA Warp, torch.compile.

Foundational: Mathematics (differential equations, linear algebra, advanced calculus), data analysis, probability and statistics, fundamental physics, mathematical modeling, research skills.

Communication: Technical and popular writing; creator of lectures and workshops on AI and scientific computing; management and mentorship of staff; teamwork and collaboration.

Experience

NVIDIA Corporation Dec 2024 – Present
Senior Software Engineer, PhysicsNeMo

Domain Parallelism: Invented and architected ShardTensor, a novel domain-parallelism framework that scales high-resolution scientific data beyond single-GPU memory limits during training and inference. Now a core capability of PhysicsNeMo. Documentation, Examples, Publication.
Performance: Served as performance lead for PhysicsNeMo, delivering up to 30x speedups on top CAE models (Blog post). Architected SciML datapipes for high-concurrency IO with GPU stream parallelism.
GeoTransolver: Authored GeoTransolver (jointly with R. Ranade), PhysicsNeMo’s top-performing surrogate model for CFD and CAE applications. Training Recipe, Publication.
v2.0 Leadership: Led the entire engineering team in the v2.0 refactor of PhysicsNeMo, delivering on time and leading to a sustained increase in downloads. Monthly downloads increased by 3x since the v2.0 release.
Top contributor to PhysicsNeMo (3k+ GitHub stars, 100+ contributors) over the past 12 months, owning domain parallelism, datapipes, multiple models, CI/CD, testing infrastructure, and multiple end-to-end training recipes.

Advanced Micro Devices (AMD) July 2024 – Dec 2024
Senior Member of Technical Staff

Lead for AI on the ROCm stack and Instinct GPU product line for Technical Marketing Engineering.
Developed content, demos, and technical and educational material related to AI GPU programming on MI300X GPUs.

Argonne National Laboratory Sept 2018 – July 2024
Computational Scientist, Physicist (Oct 2022 – July 2024); Asst. Computer Scientist, Asst. Physicist (Sept 2018 – Oct 2022)

Lead of AI software on the Aurora supercomputer, #2 in the world with more than 60k GPUs.
Developed multiple novel AI techniques for science, including graph neural networks that obey fundamental physical laws in JAX to solve quantum many-body physics.
Created a unique, fully differentiable simulation software in JAX where trainable parameters are based on interpretable physical quantities of experimental devices, trained with both supervised and unsupervised methods.
Interfaced with hardware vendors for benchmarking AI-for-science applications, including NVIDIA, SambaNova, and Cerebras.

Harvard University Feb 2017 – Sept 2018
Postdoctoral Research Associate

Physicist working on the MicroBooNE, SBND, and NEXT neutrino experiments.
Pioneered applications of deep learning to neutrino physics.

Education

Ph.D., Physics — Yale University May 2017
M.Phil., Physics — Yale University May 2014
M.Sci., Physics — Yale University May 2014
B.S., Mathematics — University of Rochester May 2011
B.S., Physics — University of Rochester May 2011

Sample Projects

PhysicsNeMo — Open-source AI physics software from NVIDIA. 3000+ stars on GitHub. Top contributor since joining the team in Dec. 2024, of 100+ contributors. Lead architect of domain parallelism, the datapipes, and multiple training recipes, as well as the core architect for the repository.
CosmicTagger — Deep-learning image segmentation (UResNet) for fundamental physics detectors. Beats state-of-the-art by 5x accuracy; deployed as a comparative benchmark and acceptance test on multiple supercomputers. PyTorch, JAX, and TF implementations. Scaled to thousands of ranks on multiple HPC systems. (Lead Developer, Inactive)
Larcv3 — C/C++ data serialization package for sparse and irregular data, built on HDF5 with a Python interface. Scaled to 20k+ ranks with MPI. (Lead Developer, Inactive)
SparseEventID — Sparse, 3D computer vision models for fundamental physics detectors. Classification as well as object detection and SimCLR unsupervised representation learning. PyTorch. (Lead Developer, Inactive)
QMC — Differentiable solver of the multibody Schrödinger equation (quantum mechanics) in JAX. Scales with mpi4jax to thousands of ranks. (Lead Developer, Inactive)

Awards

Department of Energy Early Career Award ($2.5M) — Office of Science, Nuclear Physics 2021
Impact Argonne Awards (4) — Argonne National Laboratory 2018–2022
Neutrino Physics Center Scholar — Fermi National Accelerator Laboratory 2016
University Research Association Visiting Scholar — Fermi National Accelerator Laboratory 2015
Excellence in Undergraduate Teaching Award — University of Rochester 2011

Selected Publications

Google Scholar (70+ more, h-index 40)

“GeoTransolver: Learning Physics on Irregular Domains Using Multi-scale Geometry Aware Physics Attention Transformer,” C. Adams, R. Ranade, R. Cherukuri, S. Choudhry, Submitted to NeurIPS 2026.
“ShardTensor: Domain Parallelism for Scientific Machine Learning,” C. Adams, P. Harrington, A. Subramaniam, M. S. Abbas, J. Pathak, M. Pritchard, S. Choudhry, Submitted to SuperComputing 2026.
“Demonstration of Portable Performance of Scientific Machine Learning on High Performance Computing Systems,” K. Hossain, R. Balin, C. Adams, et al., Proceedings of the SC’23 Workshops (2023).
“Nuclei with up to A = 6 nucleons with artificial neural network wave functions,” A. Gnech, C. Adams, N. Brawand, G. Carleo, A. Lovato, N. Rocco, Few-Body Systems, 63, 1 (2022).
“Variational Monte Carlo calculations of A ≤ 4 nuclei with an artificial neural-network correlator ansatz,” C. Adams, G. Carleo, A. Lovato, N. Rocco, Phys. Rev. Lett. 127, 022502 (2021).
“Demonstration of background rejection using deep convolutional neural networks in the NEXT experiment,” NEXT Collaboration, Journal of High Energy Physics, vol. 2021, art. 189 (2021).
“Cosmic Background Removal with Deep Neural Networks in SBND,” SBND Collaboration, Front. Artif. Intell. (24 August 2021).
“Convolutional Neural Networks Applied to Neutrino Events in a Liquid Argon Time Projection Chamber,” MicroBooNE Collaboration, JINST 12, P03011 (2017).

Selected Workshops, Lectures and Seminars

AI Frameworks on the Aurora Supercomputer — ALCF Developer Session, Virtual, 2024
AI for Fundamental Physics — DOE National Science Bowl, Science Day, Leesburg, VA, 2024
The Aurora Supercomputer — Argonne Colloquium, Lemont, IL, 2024
Neutrinoless Double Beta Decay at Argonne — Argonne Seminar, Lemont, IL, 2024
CNNs and GNNs for Physics — COFI Winter School, San Juan, PR, 2023
AI in the NEXT Experiment — AI in Nuclear Physics, DNP 2023
Calculation of Nuclear Ground States using Artificial Neural Networks — ALCF Hands-on Workshop, Lemont, IL, 2023
AI and Neutrino Physics — U. Chicago Physics Seminar, Hyde Park, IL, 2023
Sparse IO for Deep Learning — PyHEP Workshop, Virtual, 2022
Machine Learning Quantum Mechanics — Argonne Training Program for Extreme Scale Computing, St. Charles, IL, 2022
AI/ML on High Performance Computing — Beam Dynamics Workshop, Chicago, IL, 2022
AI in Neutrinoless Double Beta Decay — Physics Seminar, Rice University, 2022
AI and Neutrinoless Double Beta Decay — MODE Conference, Virtual, 2021
Artificial Intelligence – Computing Requirements — National Nuclear Security Administration, Emerging Technologies, Virtual, 2021
Machine Learning for Neutrinoless Double Beta Decay in NEXT — Dark-matter and Neutrino Computation Explored (DANCE) ML, Virtual, 2020
Machine Learning Day, “Autoencoders” — Argonne Training Program for Extreme Scale Computing, Virtual, 2020
Python on HPC — Simulation, Data, and Learning Workshop, Virtual, 2020
ALCF Data and Learning Frameworks — Computational Performance Workshop, Lemont, IL, 2019
Sparse Convolutional Networks — Greater Chicago Area Systems Research Workshop, Chicago, IL, 2019
Machine Learning, Data Science and Neutrino Physics at Argonne’s Leadership Computing Facility — SLAC AI Seminar, Menlo Park, CA, 2019
Deep Learning and Neutrino Physics — Argonne Leadership Computing Facility Seminar, Lemont, IL, 2018

Professional Service

Committee Service: Argonne Named Fellow Committee (2019, member; 2020–2023, co-chair); ASRC Integrated Research Infrastructure Task Force (2020–2021); ALCF Postdoc Committee (chair, 2020–2024); Argonne Postdoc Award Committee (member, 2020); Physics DEIA Committee (member, 2019–2024).

Workshops Led or Co-Organized: ALCF AI Training Series (2022, 2023); Computational Performance Workshop (2021); Neutrino Physics and Machine Learning (2020); Simulation, Data, Learning Workshop (2019).