arXiv subjects

Thorsten Kurth

Publications and source records attributed to Thorsten Kurth.

At least 37 records · Page 2Linked to original sources

arXiv2023

Spherical Fourier Neural Operators: Learning Stable Dynamics on the Sphere

Fourier Neural Operators (FNOs) have proven to be an efficient and effective method for resolution-independent operator learning in a broad variety of application areas across scientific machine learning. A key reason for their success is their ability to accurately model long-range dependencies in spatio-temporal data by learning global convolutions in a computationally efficient manner. To this end, FNOs rely on the discrete Fourier transform (DFT), however, DFTs cause visual and spectral artifacts as well as pronounced dissipation when learning operators in spherical coordinates since they incorrectly assume a flat geometry. To overcome this limitation, we generalize FNOs on the sphere, introducing Spherical FNOs (SFNOs) for learning operators on spherical geometries. We apply SFNOs to forecasting atmospheric dynamics, and demonstrate stable auto\-regressive rollouts for a year of simulated time (1,460 steps), while retaining physically plausible dynamics. The SFNO has important implications for machine learning-based simulation of climate dynamics that could eventually help accelerate our response to climate change.

cs.LG↗

arXiv2022

FourCastNet: Accelerating Global High-Resolution Weather Forecasting using Adaptive Fourier Neural Operators

Extreme weather amplified by climate change is causing increasingly devastating impacts across the globe. The current use of physics-based numerical weather prediction (NWP) limits accuracy due to high computational cost and strict time-to-solution limits. We report that a data-driven deep learning Earth system emulator, FourCastNet, can predict global weather and generate medium-range forecasts five orders-of-magnitude faster than NWP while approaching state-of-the-art accuracy. FourCast-Net is optimized and scales efficiently on three supercomputing systems: Selene, Perlmutter, and JUWELS Booster up to 3,808 NVIDIA A100 GPUs, attaining 140.8 petaFLOPS in mixed precision (11.9%of peak at that scale). The time-to-solution for training FourCastNet measured on JUWELS Booster on 3,072GPUs is 67.4minutes, resulting in an 80,000times faster time-to-solution relative to state-of-the-art NWP, in inference. FourCastNet produces accurate instantaneous weather predictions for a week in advance, enables enormous ensembles that better capture weather extremes, and supports higher global forecast resolutions.

physics.ao-ph↗

arXiv2022

FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

FourCastNet, short for Fourier Forecasting Neural Network, is a global data-driven weather forecasting model that provides accurate short to medium-range global predictions at $0.25^{\circ}$ resolution. FourCastNet accurately forecasts high-resolution, fast-timescale variables such as the surface wind speed, precipitation, and atmospheric water vapor. It has important implications for planning wind energy resources, predicting extreme weather events such as tropical cyclones, extra-tropical cyclones, and atmospheric rivers. FourCastNet matches the forecasting accuracy of the ECMWF Integrated Forecasting System (IFS), a state-of-the-art Numerical Weather Prediction (NWP) model, at short lead times for large-scale variables, while outperforming IFS for variables with complex fine-scale structure, including precipitation. FourCastNet generates a week-long forecast in less than 2 seconds, orders of magnitude faster than IFS. The speed of FourCastNet enables the creation of rapid and inexpensive large-ensemble forecasts with thousands of ensemble-members for improving probabilistic forecasting. We discuss how data-driven deep learning models such as FourCastNet are a valuable addition to the meteorology toolkit to aid and augment NWP models.

physics.ao-ph↗

arXiv2021

MLPerf HPC: A Holistic Benchmark Suite for Scientific Machine Learning on HPC Systems

Scientific communities are increasingly adopting machine learning and deep learning models in their applications to accelerate scientific insights. High performance computing systems are pushing the frontiers of performance with a rich diversity of hardware resources and massive scale-out capabilities. There is a critical need to understand fair and effective benchmarking of machine learning applications that are representative of real-world scientific use cases. MLPerf is a community-driven standard to benchmark machine learning workloads, focusing on end-to-end performance metrics. In this paper, we introduce MLPerf HPC, a benchmark suite of large-scale scientific machine learning training applications driven by the MLCommons Association. We present the results from the first submission round, including a diverse set of some of the world's largest HPC systems. We develop a systematic framework for their joint analysis and compare them in terms of data staging, algorithmic convergence, and compute performance. As a result, we gain a quantitative understanding of optimizations on different subsystems such as staging and on-node loading of data, compute-unit utilization, and communication scheduling, enabling overall $>10 \times$ (end-to-end) performance improvements through system scaling. Notably, our analysis shows a scale-dependent interplay between the dataset size, a system's memory hierarchy, and training convergence that underlines the importance of near-compute storage. To overcome the data-parallel scalability challenge at large batch sizes, we discuss specific learning techniques and hybrid data-and-model parallelism that are effective on large systems. We conclude by characterizing each benchmark with respect to low-level memory, I/O, and network behavior to parameterize extended roofline performance models in future rounds.

cs.LG↗

arXiv2020

Hierarchical Roofline Performance Analysis for Deep Learning Applications

This paper presents a practical methodology for collecting performance data necessary to conduct hierarchical Roofline analysis on NVIDIA GPUs. It discusses the extension of the Empirical Roofline Toolkit for broader support of a range of data precisions and Tensor Core support and introduces a Nsight Compute based method to accurately collect application performance information. This methodology allows for automated machine characterization and application characterization for Roofline analysis across the entire memory hierarchy on NVIDIA GPUs, and it is validated by a complex deep learning application used for climate image segmentation. We use two versions of the code, in TensorFlow and PyTorch respectively, to demonstrate the use and effectiveness of this methodology. We highlight how the application utilizes the compute and memory capabilities on the GPU and how the implementation and performance differ in two deep learning frameworks.

cs.DC↗

arXiv2020

IMPECCABLE: Integrated Modeling PipelinE for COVID Cure by Assessing Better LEads

The drug discovery process currently employed in the pharmaceutical industry typically requires about 10 years and $2-3 billion to deliver one new drug. This is both too expensive and too slow, especially in emergencies like the COVID-19 pandemic. In silicomethodologies need to be improved to better select lead compounds that can proceed to later stages of the drug discovery protocol accelerating the entire process. No single methodological approach can achieve the necessary accuracy with required efficiency. Here we describe multiple algorithmic innovations to overcome this fundamental limitation, development and deployment of computational infrastructure at scale integrates multiple artificial intelligence and simulation-based approaches. Three measures of performance are:(i) throughput, the number of ligands per unit time; (ii) scientific performance, the number of effective ligands sampled per unit time and (iii) peak performance, in flop/s. The capabilities outlined here have been used in production for several months as the workhorse of the computational infrastructure to support the capabilities of the US-DOE National Virtual Biotechnology Laboratory in combination with resources from the EU Centre of Excellence in Computational Biomedicine.

cs.DC↗

arXiv2020

Using Machine Learning to Augment Coarse-Grid Computational Fluid Dynamics Simulations

Simulation of turbulent flows at high Reynolds number is a computationally challenging task relevant to a large number of engineering and scientific applications in diverse fields such as climate science, aerodynamics, and combustion. Turbulent flows are typically modeled by the Navier-Stokes equations. Direct Numerical Simulation (DNS) of the Navier-Stokes equations with sufficient numerical resolution to capture all the relevant scales of the turbulent motions can be prohibitively expensive. Simulation at lower-resolution on a coarse-grid introduces significant errors. We introduce a machine learning (ML) technique based on a deep neural network architecture that corrects the numerical errors induced by a coarse-grid simulation of turbulent flows at high-Reynolds numbers, while simultaneously recovering an estimate of the high-resolution fields. Our proposed simulation strategy is a hybrid ML-PDE solver that is capable of obtaining a meaningful high-resolution solution trajectory while solving the system PDE at a lower resolution. The approach has the potential to dramatically reduce the expense of turbulent flow simulations. As a proof-of-concept, we demonstrate our ML-PDE strategy on a two-dimensional turbulent (Rayleigh Number $Ra=10^9$) Rayleigh-Bénard Convection (RBC) problem.

physics.comp-ph↗

arXiv2020

Time-Based Roofline for Deep Learning Performance Analysis

Deep learning applications are usually very compute-intensive and require a long run time for training and inference. This has been tackled by researchers from both hardware and software sides, and in this paper, we propose a Roofline-based approach to performance analysis to facilitate the optimization of these applications. This approach is an extension of the Roofline model widely used in traditional high-performance computing applications, and it incorporates both compute/bandwidth complexity and run time in its formulae to provide insights into deep learning-specific characteristics. We take two sets of representative kernels, 2D convolution and long short-term memory, to validate and demonstrate the use of this new approach, and investigate how arithmetic intensity, cache locality, auto-tuning, kernel launch overhead, and Tensor Core usage can affect performance. Compared to the common ad-hoc approach, this study helps form a more systematic way to analyze code performance and identify optimization opportunities for deep learning applications.

cs.DC↗

arXiv2020

$F_K / F_π$ from Möbius domain-wall fermions solved on gradient-flowed HISQ ensembles

We report the results of a lattice quantum chromodynamics calculation of $F_K/F_π$ using Möbius domain-wall fermions computed on gradient-flowed $N_f=2+1+1$ highly-improved staggered quark (HISQ) ensembles. The calculation is performed with five values of the pion mass ranging from $130 \lesssim m_π\lesssim 400$ MeV, four lattice spacings of $a\sim 0.15, 0.12, 0.09$ and $0.06$ fm and multiple values of the lattice volume. The interpolation/extrapolation to the physical pion and kaon mass point, the continuum, and infinite volume limits are performed with a variety of different extrapolation functions utilizing both the relevant mixed-action effective field theory expressions as well as discretization-enhanced continuum chiral perturbation theory formulas. We find that the $a\sim0.06$ fm ensemble is helpful, but not necessary to achieve a subpercent determination of $F_K/F_π$. We also include an estimate of the strong isospin breaking corrections and arrive at a final result of $F_{K^\pm}/F_{π^\pm} = 1.1942(45)$ with all sources of statistical and systematic uncertainty included. This is consistent with the Flavour Lattice Averaging Group average value, providing an important benchmark for our lattice action. Combining our result with experimental measurements of the pion and kaon leptonic decays leads to a determination of $|V_{us}|/|V_{ud}| = 0.2311(10)$.

hep-lat↗

arXiv2019

Lattice QCD Determination of $g_A$

The nucleon axial coupling, $g_A$, is a fundamental property of protons and neutrons, dictating the strength with which the weak axial current of the Standard Model couples to nucleons, and hence, the lifetime of a free neutron. The prominence of $g_A$ in nuclear physics has made it a benchmark quantity with which to calibrate lattice QCD calculations of nucleon structure and more complex calculations of electroweak matrix elements in one and few nucleon systems. There were a number of significant challenges in determining $g_A$, notably the notorious exponentially-bad signal-to-noise problem and the requirement for hundreds of thousands of stochastic samples, that rendered this goal more difficult to obtain than originally thought. I will describe the use of an unconventional computation method, coupled with "ludicrously'" fast GPU code, access to publicly available lattice QCD configurations from MILC and access to leadership computing that have allowed these challenges to be overcome resulting in a determination of $g_A$ with 1% precision and all sources of systematic uncertainty controlled. I will discuss the implications of these results for the convergence of $SU(2)$ Chiral Perturbation theory for nucleons, as well as prospects for further improvements to $g_A$ (sub-percent precision, for which we have preliminary results) which is part of a more comprehensive application of lattice QCD to nuclear physics. This is particularly exciting in light of the new CORAL supercomputers coming online, Sierra and Summit, for which our lattice QCD codes achieve a machine-to-machine speed up over Titan of an order of magnitude.

hep-lat↗

arXiv2019

Highly-scalable, physics-informed GANs for learning solutions of stochastic PDEs

Uncertainty quantification for forward and inverse problems is a central challenge across physical and biomedical disciplines. We address this challenge for the problem of modeling subsurface flow at the Hanford Site by combining stochastic computational models with observational data using physics-informed GAN models. The geographic extent, spatial heterogeneity, and multiple correlation length scales of the Hanford Site require training a computationally intensive GAN model to thousands of dimensions. We develop a hierarchical scheme for exploiting domain parallelism, map discriminators and generators to multiple GPUs, and employ efficient communication schemes to ensure training stability and convergence. We developed a highly optimized implementation of this scheme that scales to 27,500 NVIDIA Volta GPUs and 4584 nodes on the Summit supercomputer with a 93.1% scaling efficiency, achieving peak and sustained half-precision rates of 1228 PF/s and 1207 PF/s.

physics.comp-ph↗

arXiv2019

The Stochastic Feynman-Hellmann Method

The Feynman-Hellmann method, as implemented by Bouchard et al. [1612.06963], was recently employed successfully to determine the nucleon axial charge. A limitation of the method was the restriction to a single operator and a single momentum during the computation of each "Feynman- Hellmann" propagator. By using stochastic techniques to estimate the all-to-all propagator, we relax this constraint and demonstrate the successful implementation of this new method. We show reproduction of the axial charge on a test ensemble and non-zero momentum transfer points of the axial and vector form factors.

hep-lat↗

arXiv2019

Short Range Operator Contributions to $0νββ$ decay from LQCD

The search for neutrinoless double beta decay of nuclei is believed to be one of the most promising means to search for new physics. Observation of this very rare nuclear process, which violates Lepton Number conservation, would imply the neutrino sector has a Majorana mass component and may also provide an explanation for the universe matter-antimatter asymmetry of the universe. In the case where a heavy intermediate particle is exchanged in this process, QCD contributions from short range interactions become relevant and the calculation of matrix elements with four-quark operators becomes necessary. In these proceedings we will discuss our current progress in the calculation of these four-quark operators from LQCD.

hep-lat↗

arXiv2019

Progress in Multibaryon Spectroscopy

Anchoring the nuclear interaction in QCD is a long-outstanding problem in nuclear physics. While the lattice community has made enormous progress in mesonic physics and single nucleon physics, continuum-limit physical-point multi-nucleon physics has remained out of reach. I will review CalLat's strategy for multi-nucleon spectroscopy and our latest results.

hep-lat↗

arXiv2018

Simulating the weak death of the neutron in a femtoscale universe with near-Exascale computing

The fundamental particle theory called Quantum Chromodynamics (QCD) dictates everything about protons and neutrons, from their intrinsic properties to interactions that bind them into atomic nuclei. Quantities that cannot be fully resolved through experiment, such as the neutron lifetime (whose precise value is important for the existence of light-atomic elements that make the sun shine and life possible), may be understood through numerical solutions to QCD. We directly solve QCD using Lattice Gauge Theory and calculate nuclear observables such as neutron lifetime. We have developed an improved algorithm that exponentially decreases the time-to solution and applied it on the new CORAL supercomputers, Sierra and Summit. We use run-time autotuning to distribute GPU resources, achieving 20% performance at low node count. We also developed optimal application mapping through a job manager, which allows CPU and GPU jobs to be interleaved, yielding 15% of peak performance when deployed across large fractions of CORAL.

hep-lat↗

arXiv2018

Exascale Deep Learning for Climate Analytics

We extract pixel-level masks of extreme weather patterns using variants of Tiramisu and DeepLabv3+ neural networks. We describe improvements to the software frameworks, input pipeline, and the network training algorithms necessary to efficiently scale deep learning on the Piz Daint and Summit systems. The Tiramisu network scales to 5300 P100 GPUs with a sustained throughput of 21.0 PF/s and parallel efficiency of 79.0%. DeepLabv3+ scales up to 27360 V100 GPUs with a sustained throughput of 325.8 PF/s and a parallel efficiency of 90.7% in single precision. By taking advantage of the FP16 Tensor Cores, a half-precision version of the DeepLabv3+ network achieves a peak and sustained throughput of 1.13 EF/s and 999.0 PF/s respectively.

cs.DC↗

arXiv2018

A percent-level determination of the nucleon axial coupling from Quantum Chromodynamics

The $\textit{axial coupling of the nucleon}$, $g_A$, is the strength of its coupling to the $\textit{weak}$ axial current of the Standard Model of particle physics, in much the same way as the electric charge is the strength of the coupling to the electromagnetic current. This axial coupling dictates the rate at which neutrons decay to protons, the strength of the attractive long-range force between nucleons and other features of nuclear physics. Precision tests of the Standard Model in nuclear environments require a quantitative understanding of nuclear physics rooted in Quantum Chromodynamics, a pillar of the Standard Model. The prominence of $g_A$ makes it a benchmark quantity to determine theoretically - a difficult task because quantum chromodynamics is non-perturbative, precluding known analytical methods. Lattice Quantum Chromodynamics provides a rigorous, non-perturbative definition of quantum chromodynamics that can be implemented numerically. It has been estimated that a precision of two percent would be possible by 2020 if two challenges are overcome: contamination of $g_A$ from excited states must be controlled in the calculations and statistical precision must be improved markedly. Here we report a calculation of $g_A^{QCD} = 1.271\pm0.013$, using an unconventional method inspired by the Feynman-Hellmann theorem that overcomes these challenges.

hep-lat↗

arXiv2017

Scaling GRPC Tensorflow on 512 nodes of Cori Supercomputer

We explore scaling of the standard distributed Tensorflow with GRPC primitives on up to 512 Intel Xeon Phi (KNL) nodes of Cori supercomputer with synchronous stochastic gradient descent (SGD), and identify causes of scaling inefficiency at higher node counts. To our knowledge, this is the first exploration of distributed GRPC Tensorflow scalability on a HPC supercomputer at such large scale with synchronous SGD. We studied scaling of two convolution neural networks - ResNet-50, a state-of-the-art deep network for classification with roughly 25.5 million parameters, and HEP-CNN, a shallow topology with less than 1 million parameters for common scientific usages. For ResNet-50, we achieve >80% scaling efficiency on up to 128 workers, using 32 parameter servers (PS tasks) with a steep decline down to 23% for 512 workers using 64 PS tasks. Our analysis of the efficiency drop points to low network bandwidth utilization due to combined effect of three factors. (a) Heterogeneous distributed parallelization algorithm which uses PS tasks as centralized servers for gradient averaging is suboptimal for utilizing interconnect bandwidth. (b) Load imbalance among PS tasks hinders their efficient scaling. (c) Underlying communication primitive GRPC is currently inefficient on Cori high-speed interconnect. The HEP-CNN demands less interconnect bandwidth, and shows >80% weak scaling efficiency for up to 256 nodes with only 1 PS task. Our findings are applicable to other deep learning networks. Big networks with millions of parameters stumble upon the issues discussed here. Shallower networks like HEP-CNN with relatively lower number of parameters can efficiently enjoy weak scaling even with a single parameter server.

cs.DC↗