Lars Ruthotto

Lars Ruthotto

Winship Distinguished Research Associate Professor of Mathematics and Computer Science

Emory University, Atlanta, GA, USA

Biography

I am an applied mathematician developing computational methods for machine learning and inverse problems. I am a Winship Distinguished Research Associate Professor in the Department of Mathematics and the Department of Computer Science at Emory University and a member of Emory’s Scientific Computing Group. I lead the Emory REU/RET site for Computational Mathematics for Data Science. Prior to joining Emory, I was a postdoc at the University of British Columbia and I held PhD positions at the University of Lübeck and the University of Münster.

I received an NSF CAREER award and am also supported by grants from the US Department of Energy’s Advanced Scientific Computing Research program, the Air Force Office of Scientific Research, and contracts by Sandia National Laboratories. Previously, I was supported by the US Israeli Binational Science Foundation, and the Centers for Disease Control.

Interests

  • Scientific Computing
  • Optimization
  • Machine Learning
  • Inverse Problems
  • Image Registration

Education

  • PhD in Mathematics, 2012

    University of Münster, Germany

  • Diploma in Mathematics, 2010

    University of Münster, Germany

Research

My research and teaching sit at the intersection of applied mathematics and data science, particularly deep learning and inverse problems.

In deep learning, I seek to create new insights and efficient training for continuous models based on ordinary and partial differential equations. I also develop machine learning approaches for solving high-dimensional partial differential equations and optimal control problems and am interested in connections to active and reinforcement learning.

In inverse problems, I have been working on using generative modeling for inference. Over the years, I have been working on applications in image registration and reconstruction and computational techniques including optimal experimental design, uncertainty quantification, numerical optimization, multiscale and multigrid methods, and regularization. Over the years, I had many fruitful collaborations with domain-experts from public health, geophysics, and medical imaging.

Teaching

I offer regular one-semester courses in applied mathematics as well as advanced courses, seminars, and workshops in my research area. As part of my NSF CAREER project, I have also developed a new graduate-level course on Numerical Methods for Deep Learning. I offered this class twice at Emory and gave short versions of the class at TU Berlin, University of Chemnitz, and Scuola Normale Superiore di Pisa. I also taught a short short course on Deep Generative Modeling as part of the Spring School on Data and Models at the University of South Carolina.

In the academic year 2023-24, I am offering our Honors Linear Algebra and Vector Calculus sequence. With these courses I seek to fast track some of our freshmen with the strongest mathematics background for upper level courses and research.

Selected Activities

Team

  • Deepanshu Verma Distinguished Visiting Assistant Professor, since Sept. 2021
  • Nicole Yang Distinguished Visiting Assistant Professor, since Sept. 2022
  • Kelvin Kan, Distinguished Visiting Assistant Professor, since Sept. 2023
  • Abigail Julian, Computer Science and Informatics PhD, since Jan. 2020
  • Malvern Madondo, Computer Science and Informatics PhD, since Jan. 2020
  • Xingjian Li, Computational Mathematics PhD, since Sept. 2020
  • Haley Rosso, Computational Mathematics PhD, since Jan 2022

Selected Recent Talks

Click here for more talks.

How to Train Better: Exploiting the Separability of DNNs

invited talk at the Workshop on Dynamics and Discretization at the Simons Institute for the Theory of Computing
Oct 29, 2021 10:45 AM
Elizabeth Newman, Lars Ruthotto
Slides Video

A Machine Learning Framework for Mean Field Games and Optimal Control

invited talk at the 2021 Annual Meeting of the Canadian Applied and Industrial Mathematics Society CAIMS
Jun 21, 2021 3:31 PM

Machine Learning meets Partial Differential Equations

invited plenary talk at the joint SIAM/CAIMS Annual Meeting (AN20)
Jul 6, 2020 5:00 PM
PDF Video

Selected Publications

Click here for more publications.
Elizabeth Newman, Lars Ruthotto (2022). hessQuik: Fast Hessian computation of composite functions. The Journal of Open Source Software.

PDF Code DOI

Derek Onken, Levon Nurbekyan, Xingjian Li, Samy Wu Fung, Stanley Osher, Lars Ruthotto (2022). A Neural Network Approach for High-Dimensional Optimal Control Applied to Multi-Agent Path Finding. IEEE Transactions on Control Systems Technology.

Preprint Code DOI

Kelvin Kan, François-Xavier Aubet, Tim Januschowski, Youngsuk Park, Konstantinos Benidis, Lars Ruthotto, Jan Gasthaus (2022). Multivariate Quantile Function Forecaster. 25th International Conference on Artificial Intelligence and Statistics (AISTATS).

PDF

Elizabeth Newman, Julianne Chung, Matthias Chung, Lars Ruthotto (2021). slimTrain - A Stochastic Approximation Method for Training Separable Deep Neural Networks. arxiv.org.

Preprint

Lars Ruthotto, Eldad Haber (2021). An Introduction to Deep Generative Modeling. GAMM Mitteilungen.

Preprint DOI

Lars Ruthotto, Stanley J Osher, Wuchen Li, Levon Nurbekyan, Samy Wu Fung (2020). A machine learning framework for solving high-dimensional mean field game and mean field control problems. Proceedings of the National Academy of Sciences.

Preprint Code DOI

Lars Ruthotto, Eldad Haber (2020). Deep neural networks motivated by partial differential equations. Journal of Mathematical Imaging and Vision.

Preprint PDF DOI

Eldad Haber, Lars Ruthotto (2017). Stable architectures for deep neural networks. Inverse Problems.

Preprint PDF DOI

Contact

  • lruthotto@emory.edu
  • 400 Dowman Drive, Atlanta, GA 30322
  • Office W408, located in the 4th floor in the Math and Science Center
  • By appointment (please email me)