Seth Ockerman

Name: Seth Ockerman
Pronouns: he/him/his

Institution: University of Wisconsin-Madison
Department: Computer Science

Biography:
Seth Ockerman is a Computer Science Ph.D. candidate at the University of Wisconsin-Madison who holds a bachelor’s degree from Grand Valley State University in Computer Science. His interests are the intersection between systems and ML – particularly in the context of HPC problems. He hopes to use summer 2024 to build on his past research experience and develop lasting collaborations that he can use throughout his PhD research and future career. His past research experience is in the areas of applied ML, systems, and distributed computing. Of particular note is his experience and ongoing investigation in designing parallel Bayesian search implementations and novel parallel multi-task bayesian optimization strategies in partnership with Berkeley lab and UC Berkeley. Additionally, he designed a biology-focused image analysis software which achieved a 10,000x speedup over the state of the art (published in IEEE CIBCB 2023), created a large scale image processing pipeline which predicted COVID-19 case counts based on Twitter images (published in IEEE ICMLA 2022), and improved the efficiency of CNN training on domain-specific datasets by profiling CNN training to create a gradient boosting machine which predicted neural network performance using a few epochs of training data (published in AAAI 2023).

Academic Status: PhD Student
Year in program: 1st

What courses or academic preparation have you completed to prepare for a summer internship experience?
UG = undergraduate level | GR = graduate level GR Machine Learning, GR High Performance Computing, UG Operating Systems, UG Data Structures and Algorithms , UG Artificial Intelligence , UG Applied Machine Learning, UG Computer Organization and Assembly Language, UG Programming Languages

Have you published any research or worked on research/technical projects? Yes
Where has your research been published or where have you conducted research/technical projects? At University of Wisconsin Madison, I recently joined a team investigating online pseudo-label-based finetuning of ML models in the context of battery SOC estimation. We are targeting ML-SYS 2023 for publication. At Berkeley lab and in partnership with UC Berkeley, I investigated parallel implementations of Bayesian optimization and novel Bayesian optimization multitask learning strategies. This work is ongoing and we are looking to publish in the next 3 months. In cooperation with Van Andel Medical institute, I designed BioPII – an open source biology-focused image analysis software package which accelerate spatial cluster analysis by up to 10,000x compared to the state of the art approach. This work was published as a short paper in IEEE CIBCB 2023 – https://cmte.ieee.org/cisbbtc/wpcontent/uploads/sites/172/IEEE_CIBCB_2023_paper_2015.pdf Over the course of my undergraduate degree at Grand Valley State University, I built a large scale image processing pipeline which predicted COVID-19 case counts based on Twitter images. This work was published in IEEE ICMLA 2022 – https://ieeexplore.ieee.org/document/10068950 During an REU at Ohio State, I improved the efficiency of CNN training on domain-specific datasets by profiling CNN training to create a gradient boosting machine which predicted neural network performance using a few epochs of training data. My work was published in AAAI 2023 – https://openreview.net/pdf?id=vBSUoUuAYOA.

Please describe your research/academic interests:
I am broadly interested in the intersection between machine learning and systems – especially in the context of HPC problems. I enjoy examining AI algorithms and investigating how to best deploy them on a given architecture or creating a method to deploy them in a distributed environment. For example, this last summer I investigated how to enable Bayesian optimization to make use of parallelization in the context of super computers and HPC applications.

Computational and Data Science Areas:
Applied Mathematics; Computer Science; High-Performance Computing; Machine Learning and AI

Research Synergy:
Put simply, I am interested in AI, systems, and machine learning research because I believe in using technology to improve the world around us, and I believe that the DOE labs are committed to research that promotes the public good. My interests and experience lie in designing unique system solutions to increase the efficacy and efficiency of ML and AI – particularly on HPC systems. This aligns with the DOE lab’s HPC focus, and I believe my research in particularly relevant to the DOE labs given their utilization of supercomputers such as Perlmutter.

Motivation:
I am currently pursuing a Ph.D. in Computer Science with an emphasis on designing systems for ML in the domain of HPC. Last summer, I interned at Lawrence Berkeley lab, and I thoroughly enjoyed working in the lab environment with great coworkers who are also passionate about CS-HPC research. After spending a summer in the labs, I want to create a path to a career which allows me to continue these collaborations and work in a DOE lab. By participating in this program, I would continue to gain practical experience, valuable mentorship from experienced researchers, and gain contacts with people I would love to work alongside once I finish my Ph.D.

Lightning Talk Title: Designing HPC Systems For Large Scale Machine Learning