Lars Ruthotto: "A Numerical Analysis Perspective on Deep Neural Networks"
Machine Learning for Physics and the Physics of Learning 2019
Workshop I: From Passive to Active: Generative and Reinforcement Learning with Physics
"A Numerical Analysis Perspective on Deep Neural Networks"
Lars Ruthotto - Emory University
Abstract: In this talk, I illustrate the use of numerical analysis tools for improving the effectiveness of deep learning algorithms. With a focus on deep neural networks that can be modeled as differential equations, I highlight the importance of choosing an adequate time integrator. I also compare, using a numerical example, the difference of the first-discretize-then-differentiate and the first-differentiate-then-discretize paradigms for training residual neural networks. Finally, I show that even simple (i.e., not deep) architectures can give rise to ill-conditioned learning problems.
Institute for Pure and Applied Mathematics, UCLA
September 25, 2019
For more information: http://www.ipam.ucla.edu/mlpws1
Видео Lars Ruthotto: "A Numerical Analysis Perspective on Deep Neural Networks" канала Institute for Pure & Applied Mathematics (IPAM)
Workshop I: From Passive to Active: Generative and Reinforcement Learning with Physics
"A Numerical Analysis Perspective on Deep Neural Networks"
Lars Ruthotto - Emory University
Abstract: In this talk, I illustrate the use of numerical analysis tools for improving the effectiveness of deep learning algorithms. With a focus on deep neural networks that can be modeled as differential equations, I highlight the importance of choosing an adequate time integrator. I also compare, using a numerical example, the difference of the first-discretize-then-differentiate and the first-differentiate-then-discretize paradigms for training residual neural networks. Finally, I show that even simple (i.e., not deep) architectures can give rise to ill-conditioned learning problems.
Institute for Pure and Applied Mathematics, UCLA
September 25, 2019
For more information: http://www.ipam.ucla.edu/mlpws1
Видео Lars Ruthotto: "A Numerical Analysis Perspective on Deep Neural Networks" канала Institute for Pure & Applied Mathematics (IPAM)
Показать
Комментарии отсутствуют
Информация о видео
10 октября 2019 г. 2:27:10
00:56:01
Другие видео канала
![Osbert Bastani - Interpretable Machine Learning via Program Synthesis - IPAM at UCLA](https://i.ytimg.com/vi/TX-U9qwANPQ/default.jpg)
![Jaafar El-Awady - dislocation in high thermomechanical condition in Additive Manufacturing of Alloys](https://i.ytimg.com/vi/ElfRyYaSkYM/default.jpg)
![Vikram Gavini - Fast, Accurate and Large-scale Ab-initio Calculations for Materials Modeling](https://i.ytimg.com/vi/S9N7Tl8FPPQ/default.jpg)
![Bistra Dilkina - Machine Learning for MIP Solving - IPAM at UCLA](https://i.ytimg.com/vi/hdad1if4MqU/default.jpg)
![Amit Acharya - Slow time-scale behavior of fast microscopic dynamics - IPAM at UCLA](https://i.ytimg.com/vi/_2u60cK0Ggo/default.jpg)
![Eran Rabani - Stochastic Density Functional Theory - IPAM at UCLA](https://i.ytimg.com/vi/XeOc9XfGmAw/default.jpg)
![Xavier Bresson - Learning to Untangle Genome Assembly Graphs - IPAM at UCLA](https://i.ytimg.com/vi/gevbEfJZWkI/default.jpg)
![Liming Xiong - Prediction of Microstructure Evolution in Plastically Deformed Heterogeneous Alloys](https://i.ytimg.com/vi/13XdNjFQ9jM/default.jpg)
![Jonathan Katz - Introduction to Cryptography Part 2 of 3 - IPAM at UCLA](https://i.ytimg.com/vi/-yfcTIKBPGw/default.jpg)
![Reinhold Schneider - Multi-Reference Coupled Cluster for Computation of Excited States & Tensors](https://i.ytimg.com/vi/TT1eke0bFCE/default.jpg)
![Steve White - Model reduction using localized bases and DMRG - IPAM at UCLA](https://i.ytimg.com/vi/IIK1pLvs0DA/default.jpg)
![Jack Gilbert: "Microbiome of the Built Environment"](https://i.ytimg.com/vi/8W4LxUB-XVE/default.jpg)
![John Harrison - Formalization and Automated Reasoning: A Personal and Historical Perspective](https://i.ytimg.com/vi/AaE1mV27XxA/default.jpg)
![Garnet Chan - Lattice models and ab initio descriptions of correlated materials - IPAM at UCLA](https://i.ytimg.com/vi/3C3iYkAr064/default.jpg)
![Jan Hermann - Neural-network wave functions for quantum chemistry - IPAM at UCLA](https://i.ytimg.com/vi/O-wWPtpk5Jg/default.jpg)
![Amartya Banerjee - Electronic Structure Calculations of Chiral Matter - IPAM at UCLA](https://i.ytimg.com/vi/6-VGhXcFQpE/default.jpg)
![Rose Yu - Incorporating Symmetry for Learning Spatiotemporal Dynamics - IPAM at UCLA](https://i.ytimg.com/vi/nmxYFflfUa8/default.jpg)
![Yongsoo Yang - Neural network-assisted atomic electron tomography - IPAM at UCLA](https://i.ytimg.com/vi/hhoxcczlZZ8/default.jpg)
![Margaret Murnane - Attosecond Quantum Technologies for Advanced Materials Metrologies - IPAM at UCLA](https://i.ytimg.com/vi/B8MlLd8errI/default.jpg)
![Albert Fannjiang - From Tomographic Phase Retrieval to Projection Tomography - IPAM at UCLA](https://i.ytimg.com/vi/0Dn6aHQUBCs/default.jpg)
![Thomas Swinburne - Learning uncertainty-aware models of defect kinetics at scale - IPAM at UCLA](https://i.ytimg.com/vi/yeKwfHYu9Gc/default.jpg)