Grant: Discovery Grants Program - Individual
McGill University
$48,000.00
Efficient and Approximate Differentiable Physics-based Simulations
EF 2024-25
Résumé
The objective of this proposal is to investigate efficient and approximate gradient computation of physically based simulations. The long term goals of this work is to create a fundamentally different suite of methods that can be applied where gradients are needed. For example, this includes physically based character control, digital twins where simulations are calibrated to serve as real world proxies, optimization in industrial design and architecture, and any variety of physics-in-the-loop machine learning of natural phenomena. In the short term, I will build on my recent success with adaptive simulation techniques and explore the use of model reduction through merging models and rigidification with the goal of efficiently computing approximate (yet useful) derivatives. This proposal as a whole is timely given the current trajectory of machine learning research and the opportunity to directly incorporate physics based simulations into deep neural network architectures and generative AI models. In opposition to approaches that may aim to replace simulations with neural networks, directly incorporating simulations into a neural network architecture is appealing because it preserves our deep understanding of models, materials, geometry, discretization, numerical integration, and stability. This research will be useful to a variety of people and industries, for instance in the area of training simulations for in surgery and training for specialize vehicles or mechanisms (e.g., construction cranes, mining equipment, and forestry equipment). Overall, the work is relevant to the larger computer graphics industry, which is one where Canada has been a pioneer and continues to be strong with many companies in the visual effects, games, modeling systems and software.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2024-04523
Name nom
Kry, Paul
Department département
School of Computer Science
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
48000
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
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2716
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Virtual reality and related simulations
Research subject fr
Réalité virtuelle et simulations connexes
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
Efficient and Approximate Differentiable Physics-based Simulations
Keywords
elastic solids; friction; cloth; rigid bodies; automatic differentiation; reduced models; neural networks; optimization; machine learning
Application summary
The objective of this proposal is to investigate efficient and approximate gradient computation of physically based simulations. The long term goals of this work is to create a fundamentally different suite of methods that can be applied where gradients are needed. For example, this includes physically based character control, digital twins where simulations are calibrated to serve as real world proxies, optimization in industrial design and architecture, and any variety of physics-in-the-loop machine learning of natural phenomena. In the short term, I will build on my recent success with adaptive simulation techniques and explore the use of model reduction through merging models and rigidification with the goal of efficiently computing approximate (yet useful) derivatives. This proposal as a whole is timely given the current trajectory of machine learning research and the opportunity to directly incorporate physics based simulations into deep neural network architectures and generative AI models. In opposition to approaches that may aim to replace simulations with neural networks, directly incorporating simulations into a neural network architecture is appealing because it preserves our deep understanding of models, materials, geometry, discretization, numerical integration, and stability. This research will be useful to a variety of people and industries, for instance in the area of training simulations for in surgery and training for specialize vehicles or mechanisms (e.g., construction cranes, mining equipment, and forestry equipment). Overall, the work is relevant to the larger computer graphics industry, which is one where Canada has been a pioneer and continues to be strong with many companies in the visual effects, games, modeling systems and software.