Mary Thomas
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Mary Thomas has a total of 75 co-authors. Here are the top co-authors by number of shared publications:
- Marty Charles Kandes
San Diego Supercomputer Center
5 shared publications - Robert S. Sinkovits
San Diego Supercomputer Center, USA
4 shared publications - Bryan Chin
University of California San Diego
3 shared publications - Mahidhar Tatineni
San Diego Supercomputer Center, University of California - San Diego, La Jolla, CA, USA
3 shared publications - Arunav Gupta
Halıcıoglu Data Science Institute University of California San Diego San Diego CA
2 shared publications - Khai Vu
Computer Science and Engineering, University of California, San Diego, USA
2 shared publications
Latest Publications
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 37 (12) , 2654-2658
Critique of “Data Flow Lifecycles for Optimizing Workflow Coordination” by SCC Team From UC San Diego
Reproducibility remains a critical challenge in high-performance computing (HPC), where increasing system complexity and evolving software stacks often impede the validation of published results. This work, conducted as part of the Student Cluster Competition (SCC) 2024 Reproducibility Challenge at SC24 in Atlanta, targets this issue by reproducing key findings from Data Flow Lifecycles for Optimizing Workflow Coordination. The target paper introduces Data Flow Lifecycles (DFL), a methodology for optimizing data-intensive scientific workflows. We replicate the original experiments on SCC24 competition hardware with AMD EPYC 9634 CPU, utilizing RAM-disk and SSD-based staging. Our results successfully reproduce the original qualitative trends and demonstrate a $2.0\times$ performance speedup with these optimizations. However, quantitative results diverge by 50–62.5%, primarily due to differences in cluster architecture and system capabilities. We analyze these sources of variation and discuss implications for reproducibility in portable, power-constrained HPC environments. Our study offers practical insights for improving transparency and reproducibility in data-intensive scientific computing.
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 37 (12) , 2654-2658
Critique of “Data Flow Lifecycles for Optimizing Workflow Coordination” by SCC Team From UC San Diego
Reproducibility remains a critical challenge in high-performance computing (HPC), where increasing system complexity and evolving software stacks often impede the validation of published results. This work, conducted as part of the Student Cluster Competition (SCC) 2024 Reproducibility Challenge at SC24 in Atlanta, targets this issue by reproducing key findings from Data Flow Lifecycles for Optimizing Workflow Coordination. The target paper introduces Data Flow Lifecycles (DFL), a methodology for optimizing data-intensive scientific workflows. We replicate the original experiments on SCC24 competition hardware with AMD EPYC 9634 CPU, utilizing RAM-disk and SSD-based staging. Our results successfully reproduce the original qualitative trends and demonstrate a $2.0\times$ performance speedup with these optimizations. However, quantitative results diverge by 50–62.5%, primarily due to differences in cluster architecture and system capabilities. We analyze these sources of variation and discuss implications for reproducibility in portable, power-constrained HPC environments. Our study offers practical insights for improving transparency and reproducibility in data-intensive scientific computing.
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Bridging the Cyberinfrastructure Training Gap: Assessing the Impact of the COMPLECS HPC Training Program for Research End-Users
The COMPLECS CyberTraining program addresses the gap between traditional developer-centric high-performance computing training and the practical needs of modern researchers. Using a three-tier curriculum, the program has impacted more than 1,700 participants in its first 30 months, hosting 43 webinars and 5 immersive, multi-day workshops. Here, we review the program’s growth and offer insights for cyberinfrastructure training and improved strategies for user engagement and resource dissemination.
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Open AccessBridging the Cyberinfrastructure Training Gap: Assessing the Impact of the COMPLECS HPC Training Program for Research End-Users
The COMPLECS CyberTraining program addresses the gap between traditional developer-centric high-performance computing training and the practical needs of modern researchers. Using a three-tier curriculum, the program has impacted more than 1,700 participants in its first 30 months, hosting 43 webinars and 5 immersive, multi-day workshops. Here, we review the program’s growth and offer insights for cyberinfrastructure training and improved strategies for user engagement and resource dissemination.
- Open Access
- 17 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Preliminary Results of the MLPerf BERT Inference Benchmark on AMD Instinct GPUs
In recent years, Artificial Intelligence (AI) has reshaped various facets of our day-to-day lives. To evaluate both hardware and software deployment of Machine Learning (ML) applications, it is necessary to measure system-wide performance and accuracy. MLPerf, a Machine Learning Benchmark from MLCommons, provides a standard for measuring Machine Learning Performance on hardware. To our knowledge, there is little known public effort on porting and optimizing these inference benchmark models on AMD GPUs. This paper focuses on the effort to port MLPerf BERT, one of the MLPerf Inference Benchmarks, to run on AMD Instinct MI210 and MI250 Accelerators. We describe the challenges encountered, solutions applied, and the successful porting of the benchmark resulting in submission of the code to the MLPerf community codebase. This paper derives from unpublished work completed as part of the 2023 Student Cluster Competition (SCC23). We present the first unofficial results for running the MLPerf BERT Inference Benchmark using our optimization strategies on AMD accelerators, specifically the MI250 and MI210. Additionally, as a result of this effort, the CM Automation framework now supports AMD ROCm for PyTorch and ONNX Runtime platforms.
- 17 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessPreliminary Results of the MLPerf BERT Inference Benchmark on AMD Instinct GPUs
In recent years, Artificial Intelligence (AI) has reshaped various facets of our day-to-day lives. To evaluate both hardware and software deployment of Machine Learning (ML) applications, it is necessary to measure system-wide performance and accuracy. MLPerf, a Machine Learning Benchmark from MLCommons, provides a standard for measuring Machine Learning Performance on hardware. To our knowledge, there is little known public effort on porting and optimizing these inference benchmark models on AMD GPUs. This paper focuses on the effort to port MLPerf BERT, one of the MLPerf Inference Benchmarks, to run on AMD Instinct MI210 and MI250 Accelerators. We describe the challenges encountered, solutions applied, and the successful porting of the benchmark resulting in submission of the code to the MLPerf community codebase. This paper derives from unpublished work completed as part of the 2023 Student Cluster Competition (SCC23). We present the first unofficial results for running the MLPerf BERT Inference Benchmark using our optimization strategies on AMD accelerators, specifically the MI250 and MI210. Additionally, as a result of this effort, the CM Automation framework now supports AMD ROCm for PyTorch and ONNX Runtime platforms.
- Open Access
- 17 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Building a Federated Catalog for CyberTraining Materials: The HPC-ED Pilot Project
To improve the sharing and discovery of CyberTraining materials, the HPC-ED Pilot project team is building a platform for the community to better share and find training materials through a federated catalog. The platform, currently in early test mode, is focused on a flexible platform, informative metadata, and community participation. By creating a framework for identifying, sharing, and including content broadly, HPC-ED will: allow providers of training materials to reach new groups of learners; extend the breadth and depth of training materials; and enable local sites to add or extend local portals.
- 17 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessBuilding a Federated Catalog for CyberTraining Materials: The HPC-ED Pilot Project
To improve the sharing and discovery of CyberTraining materials, the HPC-ED Pilot project team is building a platform for the community to better share and find training materials through a federated catalog. The platform, currently in early test mode, is focused on a flexible platform, informative metadata, and community participation. By creating a framework for identifying, sharing, and including content broadly, HPC-ED will: allow providers of training materials to reach new groups of learners; extend the breadth and depth of training materials; and enable local sites to add or extend local portals.
- Open Access
- 17 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-4
COMPLECS: COMPrehensive Learning for end-users to Effectively utilize CyberinfraStructure
The needs of cyberinfrastructure (CI) Users are different from those of CI Contributors. Typically, much of the training in advanced CI addresses developer topics such as MPI, OpenMP, CUDA and application profiling, leaving a gap in training for these users. To remedy this situation, we developed a new program: COMPrehensive Learning for end-users to Effectively utilize CyberinfraStructure (COMPLECS). COMPLECS focuses exclusively on helping CI Users acquire the skills and knowledge they need to efficiently accomplish their compute- and data-intensive research, covering topics such as parallel computing concepts, data management, batch computing, cybersecurity, HPC hardware overview, and high throughput computing.
- 17 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Open AccessCOMPLECS: COMPrehensive Learning for end-users to Effectively utilize CyberinfraStructure
The needs of cyberinfrastructure (CI) Users are different from those of CI Contributors. Typically, much of the training in advanced CI addresses developer topics such as MPI, OpenMP, CUDA and application profiling, leaving a gap in training for these users. To remedy this situation, we developed a new program: COMPrehensive Learning for end-users to Effectively utilize CyberinfraStructure (COMPLECS). COMPLECS focuses exclusively on helping CI Users acquire the skills and knowledge they need to efficiently accomplish their compute- and data-intensive research, covering topics such as parallel computing concepts, data management, batch computing, cybersecurity, HPC hardware overview, and high throughput computing.
- 12 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 36 (5) , 841-846
Reproducibility of the DaCe Framework on NPBench Benchmarks
DaCe is a framework for Python that claims to provide massive speedups with C-like speeds compared to already existing high-performance Python frameworks (e.g. Numba or Pythran). In this work, we take a closer look at reproducing the NPBench work. We use performance results to confirm that NPBench achieves higher performance than NumPy in a variety of benchmarks and provide reasons as to why DaCe is not truly as portable as it claims to be, but with a small adjustment it can run anywhere.
- 12 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 36 (5) , 841-846
Reproducibility of the DaCe Framework on NPBench Benchmarks
DaCe is a framework for Python that claims to provide massive speedups with C-like speeds compared to already existing high-performance Python frameworks (e.g. Numba or Pythran). In this work, we take a closer look at reproducing the NPBench work. We use performance results to confirm that NPBench achieves higher performance than NumPy in a variety of benchmarks and provide reasons as to why DaCe is not truly as portable as it claims to be, but with a small adjustment it can run anywhere.
- Open Access
- 23 July 2023
- Published by Association for Computing Machinery (ACM)
- p. 390-394
Developing a Best Practices Training Program in Cyberinfrastructure-Enabled Machine Learning Research
- 23 July 2023
- Published by Association for Computing Machinery (ACM)
- p. 390-394
Open AccessDeveloping a Best Practices Training Program in Cyberinfrastructure-Enabled Machine Learning Research
- Open Access
- 31 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 34 (6) , 1727-1730
Critique of: “A Parallel Framework for Constraint-Based Bayesian Network Learning via Markov Blanket Discovery” by SCC Team From UC San Diego
Bayesian networks (BNs) have become popular in recent years to describe natural phenomena in situations where causal linkages are important to understand. In order to get around the inherent non-tractability of learning BNs, Srivastava et al. propose a markov blanket discovery-based approach to learning in their paper titled ”A Parallel Framework for Constraint-based Bayesian Network Learning via Markov Blanket Discovery.” We are able to reproduce both the strong and weak scaling experiments from the paper up to 128 cores, and verify communication cost scaling for all three algorithms in the paper. We also introduce methodological improvements to weak scaling that show the paper's findings are unique to the methodology and not the datasets used. Slight variations in performance were observed due to differences in datasets, core count, and job scheduling.
- 31 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 34 (6) , 1727-1730
Open AccessCritique of: “A Parallel Framework for Constraint-Based Bayesian Network Learning via Markov Blanket Discovery” by SCC Team From UC San Diego
Bayesian networks (BNs) have become popular in recent years to describe natural phenomena in situations where causal linkages are important to understand. In order to get around the inherent non-tractability of learning BNs, Srivastava et al. propose a markov blanket discovery-based approach to learning in their paper titled ”A Parallel Framework for Constraint-based Bayesian Network Learning via Markov Blanket Discovery.” We are able to reproduce both the strong and weak scaling experiments from the paper up to 128 cores, and verify communication cost scaling for all three algorithms in the paper. We also introduce methodological improvements to weak scaling that show the paper's findings are unique to the methodology and not the datasets used. Slight variations in performance were observed due to differences in datasets, core count, and job scheduling.
- 17 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 33 (9) , 2043-2046
Critique of “MemXCT: Memory-Centric X-Ray CT Reconstruction With Massive Parallelization” by SCC Team From University of California San Diego
In this article, we describe our efforts to reproduce results reported in the SC19 article by Hidayetoğlu et al. , titled “MemXCT: Memory-Centric X-ray CT Reconstruction with Massive Parallelization” . MemXCT 's single-device performance, parallelized via OpenMP and MPI, was characterized using AMD Zen2 CPU cores and NVIDIA V100 GPU devices running on the Microsoft Azure cloud. We were able to reproduce most of the results, and exceed the performance of larger inputs, on an AMD EPYC HBv2 cluster. We were also able to reproduce the strong scaling trends for optimized CPU and GPU versions. Slight variations in performance of the CPU version were observed due to differences in the underlying hardware, input size, and number of available nodes. Digital artifacts from these experiments are available at: 10.5281/zenodo.5598108
- 17 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 33 (9) , 2043-2046
Critique of “MemXCT: Memory-Centric X-Ray CT Reconstruction With Massive Parallelization” by SCC Team From University of California San Diego
In this article, we describe our efforts to reproduce results reported in the SC19 article by Hidayetoğlu et al. , titled “MemXCT: Memory-Centric X-ray CT Reconstruction with Massive Parallelization” . MemXCT 's single-device performance, parallelized via OpenMP and MPI, was characterized using AMD Zen2 CPU cores and NVIDIA V100 GPU devices running on the Microsoft Azure cloud. We were able to reproduce most of the results, and exceed the performance of larger inputs, on an AMD EPYC HBv2 cluster. We were also able to reproduce the strong scaling trends for optimized CPU and GPU versions. Slight variations in performance of the CPU version were observed due to differences in the underlying hardware, input size, and number of available nodes. Digital artifacts from these experiments are available at: 10.5281/zenodo.5598108
- Open Access
- 17 July 2021
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Expanse: Computing without Boundaries
We describe the design motivation, architecture, deployment, and early operations of Expanse, a 5 Petaflop, heterogenous HPC system that entered production as an NSF-funded resource in December 2020 and will be operated on behalf of the national community for five years. Expanse will serve a broad range of computational science and engineering through a combination of standard batch-oriented services, and by extending the system to the broader CI ecosystem through science gateways, public cloud integration, support for high throughput computing, and composable systems. Expanse was procured, deployed, and put into production entirely during the COVID-19 pandemic, adhering to stringent public health guidelines throughout. Nevertheless, the planned production date of October 1, 2020 slipped by only two months, thanks to thorough planning, a dedicated team of technical and administrative experts, collaborative vendor partnerships, and a commitment to getting an important national computing resource to the community at a time of great need.
- 17 July 2021
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Open AccessExpanse: Computing without Boundaries
We describe the design motivation, architecture, deployment, and early operations of Expanse, a 5 Petaflop, heterogenous HPC system that entered production as an NSF-funded resource in December 2020 and will be operated on behalf of the national community for five years. Expanse will serve a broad range of computational science and engineering through a combination of standard batch-oriented services, and by extending the system to the broader CI ecosystem through science gateways, public cloud integration, support for high throughput computing, and composable systems. Expanse was procured, deployed, and put into production entirely during the COVID-19 pandemic, adhering to stringent public health guidelines throughout. Nevertheless, the planned production date of October 1, 2020 slipped by only two months, thanks to thorough planning, a dedicated team of technical and administrative experts, collaborative vendor partnerships, and a commitment to getting an important national computing resource to the community at a time of great need.