Learning to Simulate Complex Physics with Graph Networks, ICML 2020
Teaser video for our ICML2020 paper.
Paper: https://arxiv.org/abs/2002.09405
More videos at: https://sites.google.com/view/learning-to-simulate
Abstract:
Here we present a general framework for learning simulation, and provide a single model implementation that yields state-of-the-art performance across a variety of challenging physical domains, involving fluids, rigid solids, and deformable materials interacting with one another. Our framework---which we term "Graph Network-based Simulators" (GNS)---represents the state of a physical system with particles, expressed as nodes in a graph, and computes dynamics via learned message-passing. Our results show that our model can generalize from single-timestep predictions with thousands of particles during training, to different initial conditions, thousands of timesteps, and at least an order of magnitude more particles at test time. Our model was robust to hyperparameter choices across various evaluation metrics: the main determinants of long-term performance were the number of message-passing steps, and mitigating the accumulation of error by corrupting the training data with noise. Our GNS framework is the most accurate general-purpose learned physics simulator to date, and holds promise for solving a wide range of complex forward and inverse problems.
Видео Learning to Simulate Complex Physics with Graph Networks, ICML 2020 канала Tobias Pfaff
Paper: https://arxiv.org/abs/2002.09405
More videos at: https://sites.google.com/view/learning-to-simulate
Abstract:
Here we present a general framework for learning simulation, and provide a single model implementation that yields state-of-the-art performance across a variety of challenging physical domains, involving fluids, rigid solids, and deformable materials interacting with one another. Our framework---which we term "Graph Network-based Simulators" (GNS)---represents the state of a physical system with particles, expressed as nodes in a graph, and computes dynamics via learned message-passing. Our results show that our model can generalize from single-timestep predictions with thousands of particles during training, to different initial conditions, thousands of timesteps, and at least an order of magnitude more particles at test time. Our model was robust to hyperparameter choices across various evaluation metrics: the main determinants of long-term performance were the number of message-passing steps, and mitigating the accumulation of error by corrupting the training data with noise. Our GNS framework is the most accurate general-purpose learned physics simulator to date, and holds promise for solving a wide range of complex forward and inverse problems.
Видео Learning to Simulate Complex Physics with Graph Networks, ICML 2020 канала Tobias Pfaff
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
![Learning Mesh-Based Simulation with Graph Networks [ICLR 2021]](https://i.ytimg.com/vi/KfZFgSff9N8/default.jpg)
![Teaser video to our NeurIPS paper "Playing hard exploration games by watching YouTube"](https://i.ytimg.com/vi/s8ZSVfYmtpc/default.jpg)
![Learnt agent - Pitfall! (updated)](https://i.ytimg.com/vi/3XMJR7z2KMc/default.jpg)
![Learnt agent - Montezuma's Revenge (updated)](https://i.ytimg.com/vi/luDm-ieOOmA/default.jpg)
![t-SNE plot of our TDC+CMC embedding](https://i.ytimg.com/vi/RyxPAYhQ-Vo/default.jpg)
![Comparison snippet - Montezuma's Revenge](https://i.ytimg.com/vi/DhUqfLjvONY/default.jpg)
![Learnt agent - Montezuma's Revenge](https://i.ytimg.com/vi/Msy82sIfprI/default.jpg)
![Learnt agent - Pitfall!](https://i.ytimg.com/vi/Oi1pXDaTECM/default.jpg)
![Learnt agent - Private Eye](https://i.ytimg.com/vi/I5itifmdrEo/default.jpg)
![God of Grammar](https://i.ytimg.com/vi/qjJZ5bSKFDY/default.jpg)
![Repairing the zbuffer using optimization and re-projection.](https://i.ytimg.com/vi/W8kHW2J826A/default.jpg)
![Adaptive Tearing and Cracking of Thin Sheets, SIGGRAPH 2014](https://i.ytimg.com/vi/4Wl0ksysYKM/default.jpg)
![A point-and-click interface for Grim Fandango](https://i.ytimg.com/vi/r8DKMM65Jk0/default.jpg)
![Fast forward: Folding and Crumpling Adaptive Sheets, SIGGRAPH 2013](https://i.ytimg.com/vi/JKZK1q8Kklc/default.jpg)
![Folding and Crumpling Adaptive Sheets, SIGGRAPH 2013](https://i.ytimg.com/vi/1zWeYgRsMbU/default.jpg)
![Siggraph Asia 2010 fast forward: Anisotropic Turbulence](https://i.ytimg.com/vi/Wn7ELcR4eCE/default.jpg)
![Lagrangian Vortex Sheets for Animating Fluids, SIGGRAPH 2012](https://i.ytimg.com/vi/doa_njZgHXk/default.jpg)
![Synthetic Turbulence using Artificial Boundary Layers , SIGGRAPH Asia 2009](https://i.ytimg.com/vi/WifYrYUVmBI/default.jpg)
![Scalable Fluid Simulation using Anisotropic Turbulence Particles, SIGGRAPH Asia 2010](https://i.ytimg.com/vi/7cC-_-aqx18/default.jpg)