Deep learning for scientific computing: (closing) the gap between theory and practice by Ben Adcock
Ben Adcock (Simon Fraser University), "Deep learning for scientific computing: (closing) the gap between theory and practice"
https://benadcock.org/
Abstract: Deep learning is starting to be increasingly used for challenging problems in scientific computing. Theoretically, such efforts are supported by a large and growing body of literature on existence of deep neural networks with favourable approximation properties. Yet, these results often say very little about practical performance in terms of the traditional pillars of numerical analysis: accuracy, stability, sampling complexity and computational cost. In this talk, I will focus on two distinct problems in scientific computing to which deep learning is being actively applied: high-dimensional function approximation and inverse problems for imaging. In each case, I will first highlight several limitations of current approaches in terms of stability, unpredictable generalization and/or the gap between existence theory and practical performance. Then, I will showcase recent theoretical contributions that show that deep neural networks matching the performance of best-in-class schemes can be computed in both settings. This highlights the potential of deep neural networks, and sheds light on achieving robust, reliable and overall improved practical performance.
References:
B. Adcock, S. Brugiapaglia, N. Dexter & S. Moraga, Deep neural networks are effective at learning high-dimensional Hilbert-valued functions from limited data, MSML21 (in revision), 2021.
B. Adcock & N. Dexter, The gap between theory and practice in function approximation with deep neural networks, SIAM J. Math. Data Sci. (to appear), 2021.
B. Adcock & A. C. Hansen. Compressive Imaging: Structure, Sampling, Learning, CUP (in press), 2021.
V. Antun, F. Renna, C. Poon, B. Adcock & A. C. Hansen, On instabilities of deep learning in image reconstruction and the potential costs of AI, Proc. Natl. Acad. Sci. USA 117(48):30088--30095, 2020.
AAAI 2021 Spring Symposium on Combining Artificial Intelligence and Machine Learning with Physics Sciences, March 22-24, 2021 (https://sites.google.com/view/aaai-mlps)
Papers: https://sites.google.com/view/aaai-mlps/proceedings
Slides: https://sites.google.com/view/aaai-mlps/program
Видео Deep learning for scientific computing: (closing) the gap between theory and practice by Ben Adcock канала MLPS - Combining AI and ML with Physics Sciences
https://benadcock.org/
Abstract: Deep learning is starting to be increasingly used for challenging problems in scientific computing. Theoretically, such efforts are supported by a large and growing body of literature on existence of deep neural networks with favourable approximation properties. Yet, these results often say very little about practical performance in terms of the traditional pillars of numerical analysis: accuracy, stability, sampling complexity and computational cost. In this talk, I will focus on two distinct problems in scientific computing to which deep learning is being actively applied: high-dimensional function approximation and inverse problems for imaging. In each case, I will first highlight several limitations of current approaches in terms of stability, unpredictable generalization and/or the gap between existence theory and practical performance. Then, I will showcase recent theoretical contributions that show that deep neural networks matching the performance of best-in-class schemes can be computed in both settings. This highlights the potential of deep neural networks, and sheds light on achieving robust, reliable and overall improved practical performance.
References:
B. Adcock, S. Brugiapaglia, N. Dexter & S. Moraga, Deep neural networks are effective at learning high-dimensional Hilbert-valued functions from limited data, MSML21 (in revision), 2021.
B. Adcock & N. Dexter, The gap between theory and practice in function approximation with deep neural networks, SIAM J. Math. Data Sci. (to appear), 2021.
B. Adcock & A. C. Hansen. Compressive Imaging: Structure, Sampling, Learning, CUP (in press), 2021.
V. Antun, F. Renna, C. Poon, B. Adcock & A. C. Hansen, On instabilities of deep learning in image reconstruction and the potential costs of AI, Proc. Natl. Acad. Sci. USA 117(48):30088--30095, 2020.
AAAI 2021 Spring Symposium on Combining Artificial Intelligence and Machine Learning with Physics Sciences, March 22-24, 2021 (https://sites.google.com/view/aaai-mlps)
Papers: https://sites.google.com/view/aaai-mlps/proceedings
Slides: https://sites.google.com/view/aaai-mlps/program
Видео Deep learning for scientific computing: (closing) the gap between theory and practice by Ben Adcock канала MLPS - Combining AI and ML with Physics Sciences
Показать
Комментарии отсутствуют
Информация о видео
13 апреля 2021 г. 6:52:10
01:09:50
Другие видео канала
Targeted use of deep learning for physics-informed model discovery by Nathan Kutz
Data-Driven Inverse Modeling with Incomplete Observations by Kailai Xu
Nonlocal Physics-Informed Neural Networks - A Unified Framework for Nonlocal Models by Marth D'Elia
ADCME MPI: Distributed Machine Learning for Computational Engineering by Kailai Xu
Accelerating Simulation of Stiff Nonlinear Systems using Continuous-Time Echo State Nets, Rackauckas
Self-Adaptive Physics-Informed Neural Networks using a Soft Attention Mechanism by Levi McClenny
Physics Informed Deep Learning for Well Test Analysis by Kamalkumar Rathinasamy
Learning High-Dimensional Hilbert-Valued Functions With DNNs From Limited Data by Nick Dexter
Combining Programmable Potentials and Neural Networks for Materials Problems by Ryan Mohr
Continuous Representation Of Molecules using Graph Variational Autoencoder by Mohammadamin Tavakoli
Permeability Prediction of Porous Media using CNN with Physical Properties by Hongkyu Yoon
TextureVAE : Learning Interpretable Representations of Material Microstructures Using VAE by Avadhut
Learning Potentials of Quantum Systems using Deep Neural Networks by Arijit Sehanobish
GMLS-Nets: A Machine Learning Framework for Unstructured Data by Nathaniel Trask
Discovery of Physics and Characterization of Microstructure with Bayesian Hidden Physics Models
Greedy Fiedler Spectral Partitioning for Data-driven Discrete Exterior Calculus by Andy Huang
Physics-Informed Spatiotemporal Deep Learning for Emulating Coupled Dynamical Systems by Diane Oyen
Accelerating high-fidelity combustion simulations with classification algorithms by Wai Tong Chung
Graph Networks with Physics-aware Knowledge Informed in Latent Space by Sungyong Seo
Graph-Informed Neural Networks by Søren Taverniers