Michael Pokojovy
Associate Professor
Department of Mathematics and Statistics
Old Dominion University
Biography
Dr. Michael Pokojovy is an Associate Professor of Data Science and School of Data Science Statistics Fellow at Old Dominion University, Norfolk, VA. His research interests include machine learning and statistics, big data analytics and scientific computing. In addition to theoretical and methodological contributions, he has a track record of applied and collaborative research in statistical process control, quantitative finance, engineering and biomedical sciences. He has authored/co-authored 40+ publications in various professional outlets and secured 10+ grants from NSF, DoEd, DHHS, DFG, etc. Dr. Pokojovy has taught various courses in data science, computational science, statistics and mathematics to undergraduate, graduate and doctoral students. He also holds an ACUE Certificate in Effective College Instruction. He has mentored and advised numerous students at all levels as well as postdoctoral scholars who are currently pursuing exciting careers in industry, government or academia. Dr. Pokojovy also serves on multiple editorial and advisory boards, in particular, he acts as an institutional partner of the Center for TAIMing AI at UNC Charlotte.
Degrees Earned
Ph.D./Mathematical Sciences/2011 Dipl.-Math/Mathematical Sciences/2007
Research Areas
Applied Mathematics; Data Science; Machine Learning/AI; other
Research Interests
My research interests are manifold. They include, but are not limited to, nonparametric and robust multivariate techniques in statistical/machine learning and AI, big data analytics, statistical process control, data mining, computational statistics, design and control of stochastic spatiotemporal systems, functional data analysis, etc. I have many years of successful experience with data science and machine learning, statistical and scientific computing (in particular, HPC) and programming in R, Matlab, Python, C/C++, etc. Also, I have a strong background in various areas of applied mathematics (such as optimization and control, (stochastic) partial differential equations, numerical analysis, etc.)
Topical Areas
Applied Computer Science; Applied Mathematics; Artificial Intelligence and Intelligent Systems; Economics and Business; Health Sciences; Informatics, Analytics and Information Science; Statistics and Probability
Research Synergy
My research interests spans multiple domains of data science and statistics with primary focus on developing, implementing and applying new statistical and AI/ML methodologies in the intersection of multivariate, nonparametric and robust statistics to solve various real-world problems in science and technology, engineering, business and economics, healthcare analytics, economics and finance, etc. The ever-expanding ubiquity of multivariate, especially high-dimensional, datasets requires new and flexible statistical techniques that are genuinely data-driven, self-adaptive and not too sensitive to contamination, missing entries or model misspecifications, etc. In harmony with NAIRR's objectives, an overarching goal of my research is to address these and other challenges of significant scientific and practical importance through innovative methodological research powered by modern public and private-sector AI resources as well advances in computing, in particular, HPC. Some of my recent research experience with AI/ML is also connected with my official capacity as an institutional partner of the Center for Trustworthy AI Through Model Risk Management (TAIM), which seeks to provide a framework to ensure that the use of AI and AI-based systems is safe, controlled and managed in all applications. Similar frameworks exist in the financial sector where a solid risk management system is established to regulate, monitor and evaluate quantitative models. As a researcher with hands-on experience in model risk management, I have a unique set of skills that can progress towards NAIRR's goals of advancing AI interpretability, security and trust. I also have extensive experience with student training and curriculum development, in particular, in data science and AI/ML, including multiple NSF and DHHS/NIH funded projects. Currently, I'm also serving as a Graduate Program Director at Old Dominion University where I am commissioned with revising and modernizing our Big Data Analytics (BDA) curriculum as part of the PhD Program in Computational and Applied Mathematics. In line with NAIRR's goals, participating in SHI 2025 would help me to learn cutting edge AI/ML best practices and expand the AI workforce by training the next generation of AI researchers and educators.
Motivation
The motivation for this application has multiple facets. Old Dominion University is a research university serving the culturally vibrant, yet socioeconomically challenged Southeastern Virginia/Hampton Roads region. Various challenges lead to adverse effects that limit career opportunities and professional growth of our students. Such effects are especially detrimental to young computational and data scientists. Participation in SHI 2025 would offer an excellent opportunity to train, mentor and support young computational and data scientists from the Hampton Roads region and help them succeed and excel in becoming AI/ML professionals and future leaders that can significantly contribute to our society and help our communities. Being a new transformative NSF-sponsored endeavor, NAIRR would also provide me with an excellent opportunity to get firsthand exposure to cutting-age AI/ML tools and infrastructure as well as equip me with superior tools to better serve in my capacity of a teacher and Graduate Program Director and contribute to AI/ML workforce training and development regionally and nationally. The personal aspect to my motivation is as follows. Inspired by a success story from one of the past participants (Dr. Sharmin Abdullah - a former student of mine, presently a collaborator and assistant professor at George Mason University), my students expressed their vivid interest in pursuing this opportunity. Having previously participated in SHI (successfully selected but not able to participate due to inability to secure funding), I have had the opportunity to meet many DOE lab scientists, primarily from the Data Science and Technology (DST) Department at Berkeley Lab. I was very impressed by the quality of their research and was pleased to observe certain parallels to our research endeavors and potential collaboration opportunities. Therefore, I was happy to support my students and prepare this application. If selected for participation, we will be delighted to participate in the Workshop and explore potential summer collaboration opportunities with NAIRR project leaders.
Supervising Students Plan
As described under “Student Merit”, I have a track record of successful collaboration with my PhD students, including on-going projects and long-term collaboration plans. Our ongoing research activities revolve around AI/ML, data science and scientific computing so we should be able to quickly kick-start our research agenda taking into account valuable inputs and suggestions from NAIRR project leader/research mentor. Based on these suggestions, we intend to develop a research proposal with our mentor that we are going to pursue during our research stay. As a faculty participant, my intention is to focus on methodological and theoretical developments, while the student participant will primarily work on developing algorithms, writing code and running simulation/empirical studies. We will be having daily meetings and brainstorming sessions as to maximize synergistic effects and establish a strong coupling between the methodological and computational component of the project. The student will be able to ask questions and get guidance while practicing to be creative. We will make all attempts to maximally benefit from the research environment and computational facilities at the hosting institution as well as explore future opportunities and new horizons. Last but not least, the student will tremendously benefit from the cultural and intellectual environment at the hosting institution and develop their communication and leadership skills. At minimum, we will aim to prepare one or two journal/conference publications with the NAIRR research mentor.
Student Merit
Samit K. Ghosh, MS is a Ph.D. student of mine. He is a fourth-year Ph.D. student in our Computational and Applied Mathematics Program. Currently, he is working on non-smooth optimization techniques in robust machine learning associated with trimmed likelihood maximization subject to regularization constraints and/or penalties. Being NP-hard mixed integer/continuous optimization problems, these problems are typically tackled with the concentration or C-step. Samit was able to show the equivalence between the C-step and the well-known Frank-Wolfe algorithm establishing the (optimal) linear convergence rate. Samit and I further plan to pursue related problems in parametric and nonparametric statistics as well as deep learning in the context of robust loss functions, which have recently gained popularity in science, engineering, business, finance, etc. Samit has a solid training in mathematics and data science, advanced computing and programming skills as well as great talent and passion for research. Lokanshu P. Malur has recently joined our PhD Program and is currently working under my supervision. Prior to joining ODU, he graduated from the Master of Science Program in Financial and Enterprise Risk Management at UConn, where I supervised his capstone project. In his capstone project, Lokanshu was working on a deep learning approach to optimal control problems with a feedback controller given by a neural network. Not only was Lokanshu able to grasp the theoretical side of this problem, but successfully managed to efficiently implement it in PyTorch/TensorFlow by writing a custom automatic differentiation routine to evaluate the Jacobian associated with this problem. The results he obtained are broadly applicable to deterministic optimal control problems in physics and engineering as well as stochastic optimal control problems in business, economics and finance. Lokanshu has a solid background in in applied mathematics, probability and statistics, econometric modeling, etc. His computational and programming skills are also superb. Based on these facts, I strongly believe that both Samit and Lokanshu will greatly benefit from a summer research project as part of Sustainable Research Pathways 2025. This will be an excellent opportunity for the students to broaden their academic horizons in connection with AI/ML and collect valuable collaboration experience with DOE and/or NAIRR project leads. Participation in SHI 2025 would encourage the students to pursue an academic career and be a truly transformational experience that would also likely inspire other students from the Hampton Roads region.
Lightning Talk Title
Data-Driven Techniques for Machine Learning, Data Science and Statistics
Keywords
Optimal control; reinforcement learning; PINN; deep Pontryagin principle; Gaussian process surrogates; non-convex optimization; Frank-Wolfe gradient descent; cluster analysis; nonparametric regression