Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

University of Toronto

$34,000.00

Generative Networks, Optimal Transport and Nonlinear Harmonic Analysis

FY 2024-25

Summary
The proposed research program seeks to build on the recent breakthroughs of the PI in two distinct research areas: the mathematics of generative neural networks and nonlinear harmonic analysis for the solution of partial differential equations. The rapid recent progress on generative neural networks has yielded spectacular success on imaging tasks such as generating realistic photographs, and data augmentation in medical imaging. Recently, these disruptive new algorithms have began to be harnessed in the solution of complex scientific computing problems, such as nonlinear partial differential equations and inverse problems. Most of these techniques are still poorly understood from a theoretical perspective. One of the long term goals of the proposed research program is to develop a rigorous mathematical understanding of generative algorithms, so as to increase their efficiency and reliability. A widely used objective function for the training of generative networks is the Wasserstein-1 distance, the minimal total cost of transporting the generated probability distribution to the probability distribution of the data. In recent work with T. Milne, the PI proved a surprising theorem: the highly successful gradient penalty Wasserstein Generative Adversarial Network algorithm which was designed to approximate the Wasserstein-1 distance actually computes a congested transport cost. This established for the first time a connection between generative networks and the mathematical theory of Congested Transport problems. This branch of Optimal Transport aims to find optimal traffic paths for moving one probability density to another, taking into account congestion effects. Part of the proposed research will study the deeper underlying geometry of these problems and its dependence on the congestion cost arising in the generative algorithms. On the numerical side, we will seek to design improved generative algorithms based on these novel insights. In joint work with I. Regev and D. Tataru in a completely different area, the PI solved a major open problem on global well-posedness for the much studied defocusing Davey-Stewartson equation, arising in modeling of water waves, as well as in ferromagnetism, plasma physics, and nonlinear optics. The solution involved a proof of the Plancherel theorem for a nonlinear Fourier transform associated to these equation, itself an open problem which had been out of reach for several decades. This nonlinear Fourier transform can also be used to solve other important dispersive partial differential equations. To do so requires proving that it has additional boundedness properties in other function spaces. Part of the proposed research will be devoted to the required challenging analysis of such nonlinear Fourier transforms.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2024-06235
Name nom
Nachman, Adrian
Department département
Departments of Mathematics and Electrical & Computer Engineering
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
34000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1205
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Mathematical sciences
Area of application fr
Sciences mathématiques
Research subject code
2956
Research subject group en
Applied mathematics
Research subject group fr
Mathématiques appliquées
Research subject en
Optimisation and optimal control theory
Research subject fr
Optimisation et théories de commande optimale
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Generative Networks, Optimal Transport and Nonlinear Harmonic Analysis
Keywords
generative networks; optimal transport; nonlinear harmonic analysis; integrable dispersive equations
Application summary
The proposed research program seeks to build on the recent breakthroughs of the PI in two distinct research areas: the mathematics of generative neural networks and nonlinear harmonic analysis for the solution of partial differential equations. The rapid recent progress on generative neural networks has yielded spectacular success on imaging tasks such as generating realistic photographs, and data augmentation in medical imaging. Recently, these disruptive new algorithms have began to be harnessed in the solution of complex scientific computing problems, such as nonlinear partial differential equations and inverse problems. Most of these techniques are still poorly understood from a theoretical perspective. One of the long term goals of the proposed research program is to develop a rigorous mathematical understanding of generative algorithms, so as to increase their efficiency and reliability. A widely used objective function for the training of generative networks is the Wasserstein-1 distance, the minimal total cost of transporting the generated probability distribution to the probability distribution of the data. In recent work with T. Milne, the PI proved a surprising theorem: the highly successful gradient penalty Wasserstein Generative Adversarial Network algorithm which was designed to approximate the Wasserstein-1 distance actually computes a congested transport cost. This established for the first time a connection between generative networks and the mathematical theory of Congested Transport problems. This branch of Optimal Transport aims to find optimal traffic paths for moving one probability density to another, taking into account congestion effects. Part of the proposed research will study the deeper underlying geometry of these problems and its dependence on the congestion cost arising in the generative algorithms. On the numerical side, we will seek to design improved generative algorithms based on these novel insights. In joint work with I. Regev and D. Tataru in a completely different area, the PI solved a major open problem on global well-posedness for the much studied defocusing Davey-Stewartson equation, arising in modeling of water waves, as well as in ferromagnetism, plasma physics, and nonlinear optics. The solution involved a proof of the Plancherel theorem for a nonlinear Fourier transform associated to these equation, itself an open problem which had been out of reach for several decades. This nonlinear Fourier transform can also be used to solve other important dispersive partial differential equations. To do so requires proving that it has additional boundedness properties in other function spaces. Part of the proposed research will be devoted to the required challenging analysis of such nonlinear Fourier transforms.