Derek Onken

About

I'm Derek Onken, a data scientist and machine learning researcher. I focus on data-intensive problems predominantly in the overlap of mathematics and computer science. Ultimately, I enjoy applying the theory to develop solutions for practical problems.

(Such approaches include the following buzzwords: deep learning, neural networks, big data, statistical analysis, GPUs, numerical optimization, continuous normalizing flows, optimal control)

 

Education

• Doctor of Philosophy - Emory University
  Computer Science & Informatics track

• Master of Science - Emory University
  Computer Science

• Bachelor of Science - University of Georgia
  Majors: Mathematics and Computer Science
  Minors: Physics and Classical Culture

Research

• Machine Learning for Pharmaceutical Applications . (2021 - present)
  Designing and implementing machine learning tools for manufacturing and clinical trials use cases.
• PDE-based Machine Learning with Lars Ruthotto . (2018 - 2021)
  Applying knowledge of optimizers and partial differential equation solvers to machine learning architectures (also known as neural differential equations). More specifically, we explore higher-order schemes, applications of optimal control, and the Discretize-Optimize approach. We also are interested in developing methods for lowering computational cost of continuous normalizing flows and tractably solving high-dimensional optimal control problems.
• Parkinson's Disease Telemonitoring and Voice Analysis via Mobile App. (2017)
  Collecting and analyzing voice data, touch pressure, and rest tremor to detect Parkinson's disease symptoms via remote patient monitoring. Building machine learning classifier using motor and vocal features extracted from these patients.
• Using CDR data from H1N1 outbreak to predict future disease outbreaks (2016-2019)
  Our dataset contains cellphone data from the H1N1 outbreak in 2009-2010 and linked health data. We develop a classifier to determine/predict sickness based on human behavior changes around Date of Diagnosis. Motivation: give health officials a more real-time and accurate estimation of sick individuals as an outbreak is occuring.
• Lunar study with Juan B. Gutierrez in UGA's Biomathematics Research Group. (2014-2016)
  Analysis of 56 million natality records and sex ratio correlations.
• Comparison of Parallelization Methods with Juan B. Gutierrez in UGA's Biomathematics Research Group. (2014)
  Exploration of MPI, CUDA, and OpenCL. Application of MPI to Geneset Enrichment Analysis (GSEA).

Publications/Preprints

D Onken, L Nurbekyan, X Li, S Wu Fung, S Osher, L Ruthotto. A Neural Network Approach for High-Dimensional Optimal Control Applied to Multi-Agent Path Finding. IEEE TCST 2022.
paper   code   videos
D Onken, L Nurbekyan, X Li, S Wu Fung, S Osher, L Ruthotto. A Neural Network Approach Applied to Multi-Agent Optimal Control. European Control Conference 2021.
paper   code   videos
D Onken, S Wu Fung, X Li, L Ruthotto. OT-Flow: Fast and Accurate Continuous Normalizing Flows via Optimal Transport. AAAI Conference on Artificial Intelligence 2021.
paper   code  
Y Vigfusson*, T Karlsson*, D Onken*, et al. Cellphone Traces Reveal Infection-Associated Behavioral Change. Proceedings of the National Academy of Science 2021.   * denotes co-first author
paper   code  
D Onken and L Ruthotto. Discretize-Optimize vs. Optimize-Discretize for Time-Series Regression and Continuous Normalizing Flows. arXiv 2020.
paper   code   videos

 

Presentations

A Neural Network Approach for High-Dimensional Optimal Control @

Optimal Transport and Mean Field Games Seminar, University of South Carolina, Mar 2021, talk

Thanks to Wuchen Li for the invitation.

UCLA Applied Mathematics Seminar, Mar 2021, talk

Thanks to Levon Nurbekyan for the invitation.

Virtual Informal Systems Seminar (VISS) at Centre for Intelligent Machines (CIM) at McGill and the Groupe d'études et de Recherche en Analyse des Décisions (GERAD), Feb 2021, talk   slides   abstract

Thanks to Peter E. Caines, Aditya Mahajan, Shuang Gao, Rinel Foguen Tchuendom, and Yaroslav Salii for the invitation.

A Neural Network Approach Applied to Multi-Agent Optimal Control @

European Control Conference, Jun 2021, talk

Thanks to the ECC organizers for the opportunity.

OT-Flow: Fast and Accurate Continuous Normalizing Flows via Optimal Transport @

AAAI Conference on Artificial Intelligence, Feb 2021, talk   slides   poster

Thanks to the AAAI organizers for the opportunity.

Georgia Scientific Computing Symposium, Feb 2021, poster

Thanks to the organizers at UGA for the opportunity.

SIAM Annual Meeting, Jul 2020, poster

Thanks to the SIAM AN20 organizers for the opportunity.

Discretize-Optimize Methods for Neural ODEs in Continuous Normalizing Flows @

SIAM Mathematics of Data Science, Jun 2020, slides

Thanks to Harbir Antil, Thomas Brown, and Ratna Khatri for the invitation.

Other

Deep Learning in Medical Applications Workshop at Institute for Pure and Applied Mathematics (IPAM), Jan 2020, poster

Thanks to the IPAM organizers for the opportunity.

Emory Scientific Computing Seminar, Apr 2019, slides

Georgia Scientific Computing Symposium, Feb 2019, poster

Amazon Graduate Research Symposium, Oct 2017, poster

 

Work Experience

2021 - Present

Research Scientist for Eli Lilly

Summer 2020:

Graduate Data Science Intern for United Health Group

Summer 2019:

Graduate Data Science Intern for United Health Group

Summer 2018:

High Performance Computing Intern for Air Force Research Labs

Spring 2018:

TA for Dr. Li Xiong's Data Mining (CS378)

Fall 2017:

Lab Instructor for Dr. Lars Ruthotto's Numerical Analysis (MATH315)

Spring 2017:

TA for Dr. Michele Benzi's Numerical Analysis (MATH315)

Fall 2016:

TA for Dr. Davide Fossati's Intro to Java (CS170)

 

Contact Me

Email: derek [at] derekonken [dot] com

Other

This is my CV.
Here are some Fun Links.