Tobias Pfaff (DeepMind): Learning to Simulate Complex Physics with Graph Networks

Channel:
Canada LLMs Explained - Aggregate Intellect - AI.SCIENCE
Subscribers:
22,300
Published on ● Video Link: https://www.youtube.com/watch?v=8v27_jzNynM


Duration: 54:04
4,865 views
112

We have invited Tobias Pfaff from DeepMind to speak about his team's recent paper which presents a general framework called "Graph Network-based Simulators (GNS)" for learning physical simulation with a single model. It achieves state-of-the-art performance across a variety of challenging physical domains, involving fluids, rigid solids, and deformable materials interacting with one another.

For slides and more information on the paper, visit https://ai.science/e/learning-to-simulate-complex-physics-with-graph-networks--reO5bMSsNtcJcpr1Muvt

Speaker / coauthor: Tobias Pfaff (Research Scientist at DeepMind)

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.


AISC Spotlight



Other Videos By LLMs Explained - Aggregate Intellect - AI.SCIENCE


2020-05-22Machine Learning Methods for High Throughput Virtual Screening with a focus on Organic Photovoltaics
2020-05-21Learning the Graphical Structure of Electronic Health Records with Graph Convolutional Transformer
2020-05-20Reinforcement Learning for Batch-to-Batch Bioprocess Optimisation | AISC
2020-05-20Leaf Doctor: Plant Disease Detection Using Image Classification | Deep Learning Workshop Capstone
2020-05-20News ScanNER: Entity Tagging in News Headlines | Deep Learning Workshop Capstone
2020-05-19New methods for identifying latent manifold structure from neural data | ASIC
2020-05-19Using unsupervised machine learning to uncover hidden scientific knowledge | AISC
2020-05-15FineGym: A Hierarchical Video Dataset for Fine-grained Action Understanding
2020-05-14A Literature Review on ML in Climate Science | AISC
2020-05-13[cnvrg.io] Operating System for Machine Learning | AISC
2020-05-12Tobias Pfaff (DeepMind): Learning to Simulate Complex Physics with Graph Networks
2020-05-07Multi Type Mean Field Reinforcement Learning | AISC
2020-05-07Proving the Lottery Ticket Hypothesis: Pruning is All You Need | AISC Livestream with the Author
2020-05-06Generative Teaching Networks: Accelerating Neural Architecture Search by Learning to Generate Synthe
2020-05-06A Literature Review on Deep Learning in Finance | AISC
2020-05-02COVID19 and AI: Ethics and Data Rights Panel | AISC & NYAI
2020-04-30A Literature Review on Interpretability for Machine Learning | AISC
2020-04-29A Literature Review on ML in Health Care : Introducing new AISC Stream | AISC
2020-04-29A Literature Review on ML in Neuroscience - Introducing new AISC Stream | AISC
2020-04-28Model Selection for Optimal Prediction in Statistical Learning (Part 1: 7 Wheels of Stat Learning)
2020-04-28A Literature Review on Machine Learning in Materials Science | AISC


Tags:
graph neural networks
deep learning
deepmind
machine learning
physics simulation