Exascale Deep Learning for Climate Analytics

Kurth, Thorsten; Treichler, Sean; Romero, Joshua; Mudigonda, Mayur; Luehr, Nathan; Phillips, Everett C.; Mahesh, Ankur; Matheson, Michael J.; Deslippe, Jack; Fatica, Massimiliano; Prabhat,; Houston, Michael E.

doi:10.48550/arxiv.1810.01993

Cited by 4 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, the training process might often be accelerated by appropriate metrics for prediction error-loss function. The network complexity should also reflect available computational resources, because optimization and training are inevitably expensive in terms of computations (e.g., Gordon Bell prize 2018 for Kurth et al, 2018). In this section, we present a conceptual architecture of the neural network along with workflows to finding optimal parameters of the network.…”

Section: Predictor Philosophymentioning

confidence: 99%

Toward Automated Early Detection of Risks for a CO₂ Plume Containment From Permanent Seismic Monitoring Data

et al. 2021

View full text Add to dashboard Cite

show abstract

Section: Predictor Philosophymentioning

confidence: 99%

Toward Automated Early Detection of Risks for a CO₂ Plume Containment From Permanent Seismic Monitoring Data

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Therefore, recent hardware provide higher 32-bit and 16-bit processing rates at the expense of double precision. For example, the tensor cores on the Nvidia Volta architecture enable OLCF's Summit to already breach the exaops barrier [115]. Intel, in turn, has announced the Knights Mill architecture which is an Intel Xeon Phi chip with improved single precision performance compared to the previous Knights Landing architecture [116] designed specifically to serve the artificial intelligence market.…”

Section: Mixed Precisionmentioning

confidence: 99%

Status and future perspectives for lattice gauge theory calculations to the exascale and beyond

et al. 2019

View full text Add to dashboard Cite

In this and a set of companion whitepapers, the USQCD Collaboration lays out a program of science and computing for lattice gauge theory. These whitepapers describe how calculation using lattice QCD (and other gauge theories) can aid the interpretation of ongoing and upcoming experiments in particle and nuclear physics, as well as inspire new ones. * Editor, bjoo@jlab.org † Editor, chulwoo@bnl.gov arXiv:1904.09725v2 [hep-lat]

show abstract

“…At a cost of a relatively expensive computation in the training process, deep neural networks (DNNs) provide a powerful approach to explore hidden correlations in massive data, which in many cases are physically not possible with human manual review [2]. In the past decade, the large computational cost for training DNN has been mitigated by a number of advances, including high-performance computers [3], graphics processing units (GPUs) [4], tensor processing units (TPUs) [5], and fast largescale optimization schemes [6], i.e., adaptive moment estimation (Adam) [7] and adaptive gradient algorithm (AdaGrad) [8]. In many instances for modeling physical systems, physical invariants, e.g., momentum and energy conservation laws, can be built into the learning algorithms in the context of DNN and their variants [9][10][11].…”

Section: Introductionmentioning

confidence: 99%

PPINN: Parareal Physics-Informed Neural Network for time-dependent PDEs

Meng,

Li,

Zhang

et al. 2019

Preprint

View full text Add to dashboard Cite

Physics-informed neural networks (PINNs) encode physical conservation laws and prior physical knowledge into the neural networks, ensuring the correct physics is represented accurately while alleviating the need for supervised learning to a great degree [1]. While effective for relatively short-term time integration, when long time integration of the time-dependent PDEs is sought, the time-space domain may become arbitrarily large and hence training of the neural network may become prohibitively expensive. To this end, we develop a parareal physics-informed neural network (PPINN), hence decomposing a long-time problem into many independent short-time problems supervised by an inexpensive/fast coarse-grained (CG) solver. In particular, the serial CG solver is designed to provide approximate predictions of the solution at discrete times, while initiate many fine PINNs simultaneously to correct the solution iteratively. There is a two-fold benefit from training PINNs with small-data sets rather than working on a large-data set directly, i.e., training of individual PINNs with small-data is much faster, while training the fine PINNs can be readily parallelized. Consequently, compared to the original PINN approach, the proposed PPINN approach may achieve a significant speedup for long-time integration of PDEs, assuming that the CG solver is fast and can provide reasonable predictions of the solution, hence aiding the PPINN solution to converge in just a few iterations. To investigate the PPINN performance on solving time-dependent PDEs, we first apply the PPINN to solve the Burgers equation, and subsequently we apply the PPINN to solve a two-dimensional nonlinear diffusion-reaction equation. Our results demonstrate that PPINNs converge in a couple of iterations with significant speed-ups proportional to the number of time-subdomains employed.

show abstract

Exascale Deep Learning for Climate Analytics

Cited by 4 publications

References 0 publications

Toward Automated Early Detection of Risks for a CO₂ Plume Containment From Permanent Seismic Monitoring Data

Toward Automated Early Detection of Risks for a CO₂ Plume Containment From Permanent Seismic Monitoring Data

Status and future perspectives for lattice gauge theory calculations to the exascale and beyond

PPINN: Parareal Physics-Informed Neural Network for time-dependent PDEs

Contact Info

Product

Resources

About

Exascale Deep Learning for Climate Analytics

Cited by 4 publications

References 0 publications

Toward Automated Early Detection of Risks for a CO2 Plume Containment From Permanent Seismic Monitoring Data

Toward Automated Early Detection of Risks for a CO2 Plume Containment From Permanent Seismic Monitoring Data

Status and future perspectives for lattice gauge theory calculations to the exascale and beyond

PPINN: Parareal Physics-Informed Neural Network for time-dependent PDEs

Contact Info

Product

Resources

About

Toward Automated Early Detection of Risks for a CO₂ Plume Containment From Permanent Seismic Monitoring Data

Toward Automated Early Detection of Risks for a CO₂ Plume Containment From Permanent Seismic Monitoring Data