Lars Ruthotto

Lars Ruthotto

Auto-generated profile, based on publicly available information

Lars Ruthotto

Departments of Mathematics and Computer Science Emory University Atlanta GA USA

Research Focus & Metrics

Academic Career Length

17Years (2010-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 December 2017 in Inverse Problems
Verified
Journal Article
Closed access
PDF file
EHEldad Haber
Eldad Haber
LRLars Ruthotto
Lars Ruthotto
Deep neural networks have become invaluable tools for supervised machine learning, e.g. classification of text or images. While often offering superior results over traditional techniques and successfully expressing complicated patterns in data, deep architectures are known to be challenging to design and train such that they generalize well to new data. Critical issues with deep architectures are numerical instabilities in derivative-based learning algorithms commonly called exploding or vanishing gradients. In this paper, we propose new forward propagation techniques inspired by systems of ordinary differential equations (ODE) that overcome this challenge and lead to well-posed learning problems for arbitrarily deep networks. The backbone of our approach is our interpretation of deep learning as a parameter estimation problem of nonlinear dynamical systems. Given this formulation, we analyze stability and well-posedness of deep learning and use this new understanding to develop new network architectures. We relate the exploding and vanishing gradient phenomenon to the stability of the discrete ODE and present several strategies for stabilizing deep learning for very deep networks. While our new architectures restrict the solution space, several numerical experiments show their competitiveness with state-of-the-art networks.

18 September 2019 in Journal of Mathematical Imaging and Vision
Verified
Journal Article
Closed access
PDF file
LRLars Ruthotto
Lars Ruthotto
EHEldad Haber
Eldad Haber

9 April 2020 in Proceedings of the National Academy of Sciences
Verified
Journal Article
Closed access
PDF file
LRLars Ruthotto
Lars Ruthotto
SOStanley J. Osher
Stanley J. Osher
WLWuchen Li
Wuchen Li
LNLevon Nurbekyan
Levon Nurbekyan
SFSamy Wu Fung
Samy Wu Fung
Mean field games (MFG) and mean field control (MFC) are critical classes of multiagent models for the efficient analysis of massive populations of interacting agents. Their areas of application span topics in economics, finance, game theory, industrial engineering, crowd motion, and more. In this paper, we provide a flexible machine learning framework for the numerical solution of potential MFG and MFC models. State-of-the-art numerical methods for solving such problems utilize spatial discretization that leads to a curse of dimensionality. We approximately solve high-dimensional problems by combining Lagrangian and Eulerian viewpoints and leveraging recent advances from machine learning. More precisely, we work with a Lagrangian formulation of the problem and enforce the underlying Hamilton–Jacobi–Bellman (HJB) equation that is derived from the Eulerian formulation. Finally, a tailored neural network parameterization of the MFG/MFC solution helps us avoid any spatial discretization. Our numerical results include the approximate solution of 100-dimensional instances of optimal transport and crowd motion problems on a standard work station and a validation using a Eulerian solver in two dimensions. These results open the door to much-anticipated applications of MFG and MFC models that are beyond reach with existing numerical methods.

1 July 2019 in NeuroImage
Verified
Journal Article
Open access
KTKarsten Tabelow
Karsten Tabelow
EBEvelyne Balteau
Evelyne Balteau
JAJohn Ashburner
John Ashburner
MCMartina F. Callaghan
Martina F. Callaghan
BDBogdan Draganski
Bogdan Draganski
SMSiawoosh Mohammadi
Siawoosh Mohammadi
Neuroscience and clinical researchers are increasingly interested in quantitative magnetic resonance imaging (qMRI) due to its sensitivity to micro-structural properties of brain tissue such as axon, myelin, iron and water concentration. We introduce the hMRI-toolbox, an open-source, easy-to-use tool available on GitHub, for qMRI data handling and processing, presented together with a tutorial and example dataset. This toolbox allows the estimation of high-quality multi-parameter qMRI maps (longitudinal and effective transverse relaxation rates R1 and R2⋆, proton density PD and magnetisation transfer MT saturation) that can be used for quantitative parameter analysis and accurate delineation of subcortical brain structures. The qMRI maps generated by the toolbox are key input parameters for biophysical models designed to estimate tissue microstructure properties such as the MR g-ratio and to derive standard and novel MRI biomarkers. Thus, the current version of the toolbox is a first step towards in vivo histology using MRI (hMRI) and is being extended further in this direction. Embedded in the Statistical Parametric Mapping (SPM) framework, it benefits from the extensive range of established SPM tools for high-accuracy spatial registration and statistical inferences and can be readily combined with existing SPM toolboxes for estimating diffusion MRI parameter maps. From a user's perspective, the hMRI-toolbox is an efficient, robust and simple framework for investigating qMRI data in neuroscience and clinical research.

28 May 2021 in GAMM-Mitteilungen
Verified
Journal Article
Closed access
PDF file
LRLars Ruthotto
Lars Ruthotto
EHEldad Haber
Eldad Haber

1 January 2013 in SIAM Journal on Scientific Computing
Verified
Journal Article
Closed access
MBMartin Burger
Martin Burger
JMJan Modersitzki
Jan Modersitzki
LRLars Ruthotto
Lars Ruthotto
Image registration is one of the most challenging problems in image processing, where ill-posedness arises due to noisy data as well as nonuniqueness, and hence the choice of regularization is crucial. This paper presents hyperelasticity as a regularizer and introduces a new and stable numerical implementation. On one hand, hyperelastic registration is an appropriate model for large and highly nonlinear deformations, for which a linear elastic model needs to fail. On the other hand, the hyperelastic regularizer yields very regular and diffeomorphic transformations. While hyperelasticity might be considered as just an additional outstanding regularization option for some applications, it becomes inevitable for applications involving higher order distance measures like mass-preserving registration. The paper gives a short introduction to image registration and hyperelasticity. The hyperelastic image registration problem is phrased in a variational setting, and an existence proof is provided. The focus of the paper, however, is on a robust numerical scheme. A key challenge is an unbiased discretization of hyperelasticity, which enables the numerical monitoring of variations of length, surface, and volume of infinitesimal reference elements. We resolve this issue by using a nodal-based discretization with a special tetrahedral partitioning. The potential of the hyperelastic registration is demonstrated in a direct comparison with a linear elastic registration in an academic example. The paper also presents a real life application from three-dimensional positron emission tomography of the human heart which requires mass preservation, and thus hyperelastic registration is the only option.