Showing 1–7 of 7 results for author: Kersten, K

Empowering Federated Learning for Massive Models with NVIDIA FLARE

Authors: Holger R. Roth, Ziyue Xu, Yuan-Ting Hsieh, Adithya Renduchintala, Isaac Yang, Zhihong Zhang, Yuhong Wen, Sean Yang, Kevin Lu, Kristopher Kersten, Camir Ricketts, Daguang Xu, Chester Chen, Yan Cheng, Andrew Feng

Abstract: In the ever-evolving landscape of artificial intelligence (AI) and large language models (LLMs), handling and leveraging data effectively has become a critical challenge. Most state-of-the-art machine learning algorithms are data-centric. However, as the lifeblood of model performance, necessary data cannot always be centralized due to various factors such as privacy, regulation, geopolitics, copy… ▽ More In the ever-evolving landscape of artificial intelligence (AI) and large language models (LLMs), handling and leveraging data effectively has become a critical challenge. Most state-of-the-art machine learning algorithms are data-centric. However, as the lifeblood of model performance, necessary data cannot always be centralized due to various factors such as privacy, regulation, geopolitics, copyright issues, and the sheer effort required to move vast datasets. In this paper, we explore how federated learning enabled by NVIDIA FLARE can address these challenges with easy and scalable integration capabilities, enabling parameter-efficient and full supervised fine-tuning of LLMs for natural language processing and biopharmaceutical applications to enhance their accuracy and robustness. △ Less

Submitted 12 February, 2024; originally announced February 2024.

NVIDIA FLARE: Federated Learning from Simulation to Real-World

Authors: Holger R. Roth, Yan Cheng, Yuhong Wen, Isaac Yang, Ziyue Xu, Yuan-Ting Hsieh, Kristopher Kersten, Ahmed Harouni, Can Zhao, Kevin Lu, Zhihong Zhang, Wenqi Li, Andriy Myronenko, Dong Yang, Sean Yang, Nicola Rieke, Abood Quraini, Chester Chen, Daguang Xu, Nic Ma, Prerna Dogra, Mona Flores, Andrew Feng

Abstract: Federated learning (FL) enables building robust and generalizable AI models by leveraging diverse datasets from multiple collaborators without centralizing the data. We created NVIDIA FLARE as an open-source software development kit (SDK) to make it easier for data scientists to use FL in their research and real-world applications. The SDK includes solutions for state-of-the-art FL algorithms and… ▽ More Federated learning (FL) enables building robust and generalizable AI models by leveraging diverse datasets from multiple collaborators without centralizing the data. We created NVIDIA FLARE as an open-source software development kit (SDK) to make it easier for data scientists to use FL in their research and real-world applications. The SDK includes solutions for state-of-the-art FL algorithms and federated machine learning approaches, which facilitate building workflows for distributed learning across enterprises and enable platform developers to create a secure, privacy-preserving offering for multiparty collaboration utilizing homomorphic encryption or differential privacy. The SDK is a lightweight, flexible, and scalable Python package. It allows researchers to apply their data science workflows in any training libraries (PyTorch, TensorFlow, XGBoost, or even NumPy) in real-world FL settings. This paper introduces the key design principles of NVFlare and illustrates some use cases (e.g., COVID analysis) with customizable FL workflows that implement different privacy-preserving algorithms. Code is available at https://github.com/NVIDIA/NVFlare. △ Less

Submitted 28 April, 2023; v1 submitted 24 October, 2022; originally announced October 2022.

Comments: Accepted at the International Workshop on Federated Learning, NeurIPS 2022, New Orleans, USA (https://federated-learning.org/fl-neurips-2022); Revised version v2: added Key Components list, system metrics for homomorphic encryption experiment; Extended v3 for journal submission

Journal ref: IEEE Data Eng. Bull., Vol. 46, No. 1, 2023

The Rapid Variability of Electric Field Waves within and near interplanetary shock ramps: STEREO Observations

Authors: Z. A. Cohen, C. A. Cattell, A. W. Breneman, L. Davis, P. Grul, K. Kersten, L. B. Wilson III, J. R. Wygant

Abstract: We present STEREO observations within 1500 proton gyroradii of 12 interplanetary shocks, with long-duration burst mode electric field acquisition by S/WAVES enabling observation of the evolution of waves throughout the entire ramp of interplanetary shocks. The shocks are low Mach number ($M_{f} \sim $1--5), quasi-perpendicular ($θ_{Bn} \geq 45^{\circ}$), with beta ($β$) $\sim $0.2--1.8. High varia… ▽ More We present STEREO observations within 1500 proton gyroradii of 12 interplanetary shocks, with long-duration burst mode electric field acquisition by S/WAVES enabling observation of the evolution of waves throughout the entire ramp of interplanetary shocks. The shocks are low Mach number ($M_{f} \sim $1--5), quasi-perpendicular ($θ_{Bn} \geq 45^{\circ}$), with beta ($β$) $\sim $0.2--1.8. High variability in frequency, amplitude, and wave mode is observed upstream, downstream, and in shock ramps. Observations in every region include ion acoustic-like waves, electron cyclotron drift instability driven waves, electrostatic solitary waves, and high frequency whistler mode waves. We also show for the first time the existence of "dispersive" electrostatic waves with frequencies in the ion acoustic range and the first observations of electron cyclotron drift instability (ECDI) driven waves at interplanetary shocks. Large amplitude waves are bursty and seen in all three regions with amplitudes from $\sim$5 to $>$ 200 mV/m. All wave modes are more commonly observed downstream of the shocks than upstream of them, usually within $\sim$ 63000 km ($\sim 1500 ρ_{gi}$) of the ramp. △ Less

Submitted 7 October, 2020; v1 submitted 17 September, 2019; originally announced September 2019.

Electromagnetic waves and electron anisotropies downstream of supercritical interplanetary shocks

Authors: L. B. Wilson III, A. Koval, A. Szabo, A. Breneman, C. A. Cattell, K. Goetz, P. J. Kellogg, K. Kersten, J. C. Kasper, B. A. Maruca, M. Pulupa

Abstract: We present waveform observations of electromagnetic lower hybrid and whistler waves with f_ci << f < f_ce downstream of four supercritical interplanetary (IP) shocks using the Wind search coil magnetometer. The whistler waves were observed to have a weak positive correlation between \partialB and normalized heat flux magnitude and an inverse correlation with T_eh/T_ec. All were observed simultaneo… ▽ More We present waveform observations of electromagnetic lower hybrid and whistler waves with f_ci << f < f_ce downstream of four supercritical interplanetary (IP) shocks using the Wind search coil magnetometer. The whistler waves were observed to have a weak positive correlation between \partialB and normalized heat flux magnitude and an inverse correlation with T_eh/T_ec. All were observed simultaneous with electron distributions satisfying the whistler heat flux instability threshold and most with T_{perp,h}/T_{para,h} > 1.01. Thus, the whistler mode waves appear to be driven by a heat flux instability and cause perpendicular heating of the halo electrons. The lower hybrid waves show a much weaker correlation between \partialB and normalized heat flux magnitude and are often observed near magnetic field gradients. A third type of event shows fluctuations consistent with a mixture of both lower hybrid and whistler mode waves. These results suggest that whistler waves may indeed be regulating the electron heat flux and the halo temperature anisotropy, which is important for theories and simulations of electron distribution evolution from the sun to the earth. △ Less

Submitted 26 July, 2012; originally announced July 2012.

Comments: 20 pages, 3 PDF figures, submitted to Journal of Geophysical Research

Using an Ellipsoid Model to Track and Predict the Evolution and Propagation of Coronal Mass Ejections

Authors: Samuel Schreiner, Cynthia Cattell, Kris Kersten, Adam Hupach

Abstract: We present a method for tracking and predicting the propagation and evolution of coronal mass ejections (CMEs) using the imagers on the STEREO and SOHO satellites. By empirically modeling the material between the inner core and leading edge of a CME as an expanding, outward propagating ellipsoid, we track its evolution in three-dimensional space. Though more complex empirical CME models have been… ▽ More We present a method for tracking and predicting the propagation and evolution of coronal mass ejections (CMEs) using the imagers on the STEREO and SOHO satellites. By empirically modeling the material between the inner core and leading edge of a CME as an expanding, outward propagating ellipsoid, we track its evolution in three-dimensional space. Though more complex empirical CME models have been developed, we examine the accuracy of this relatively simple geometric model, which incorporates relatively few physical assumptions, including i) a constant propagation angle and ii) an azimuthally symmetric structure. Testing our ellipsoid model developed herein on three separate CMEs, we find that it is an effective tool for predicting the arrival of density enhancements and the duration of each event near 1 AU. For each CME studied, the trends in the trajectory, as well as the radial and transverse expansion are studied from 0 to ~.3 AU to create predictions at 1 AU with an average accuracy of 2.9 hours. △ Less

Submitted 26 February, 2012; originally announced February 2012.

Comments: 18 pages, 11 figures

Observation of relativistic electron microbursts in conjunction with intense radiation belt whistler-mode waves

Authors: K. Kersten, C. A. Cattell, A. Breneman, K. Goetz, P. J. Kellogg, L. B. Wilson III, J. R. Wygant, J. B. Blake, M. D. Looper, I. Roth

Abstract: We present multi-satellite observations indicating a strong correlation between large amplitude radiation belt whistler-mode waves and relativistic electron precipitation. On separate occasions during the Wind petal orbits and STEREO phasing orbits, Wind and STEREO recorded intense whistler-mode waves in the outer nightside equatorial radiation belt with peak-to-peak amplitudes exceeding 300 mV/m.… ▽ More We present multi-satellite observations indicating a strong correlation between large amplitude radiation belt whistler-mode waves and relativistic electron precipitation. On separate occasions during the Wind petal orbits and STEREO phasing orbits, Wind and STEREO recorded intense whistler-mode waves in the outer nightside equatorial radiation belt with peak-to-peak amplitudes exceeding 300 mV/m. During these intervals of intense wave activity, SAMPEX recorded relativistic electron microbursts in near magnetic conjunction with Wind and STEREO. The microburst precipitation exhibits a bursty temporal structure similar to that of the observed large amplitude wave packets, suggesting a connection between the two phenomena. Simulation studies corroborate this idea, showing that nonlinear wave--particle interactions may result in rapid energization and scattering on timescales comparable to those of the impulsive relativistic electron precipitation. △ Less

Submitted 4 April, 2011; v1 submitted 17 January, 2011; originally announced January 2011.

Comments: This paper has been withdrawn by the authors

A statistical study of the properties of large amplitude whistler waves and their association with few eV to 30 keV electron distributions observed in the magnetosphere by Wind

Authors: L. B. Wilson III, C. A. Cattell, P. J. Kellogg, J. R. Wygant, K. Goetz, A. Breneman, K. Kersten

Abstract: We present a statistical study of the characteristics of very large amplitude whistler waves inside the terrestrial magnetosphere using waveform capture data from the Wind spacecraft as an addition of the study by Kellogg et al., [2010b]. We observed 244(65) whistler waves using electric(magnetic) field data from the Wind spacecraft finding ~40%(~62%) of the waves have peak-to-peak amplitudes of >… ▽ More We present a statistical study of the characteristics of very large amplitude whistler waves inside the terrestrial magnetosphere using waveform capture data from the Wind spacecraft as an addition of the study by Kellogg et al., [2010b]. We observed 244(65) whistler waves using electric(magnetic) field data from the Wind spacecraft finding ~40%(~62%) of the waves have peak-to-peak amplitudes of >/- 50 mV/m(>/- 0.5 nT). We present an example waveform capture of the largest magnetic field amplitude (>/- 8 nT peak-to-peak) whistler wave ever reported in the radiation belts. The estimated Poynting flux magnitude associated with this wave is >/- 300 microW/m^2, roughly four orders of magnitude above previous estimates. Such large Poynting flux values are consistent with rapid energization of electrons. The majority of the largest amplitude whistlers occur during magnetically active periods (AE > 200 nT). The waves were observed to exhibit a broad range of propagation angles with respect to the magnetic field, 0° </- θ_kB < 90°, which showed no consistent variation with magnetic latitude. These results are inconsistent with the idea that the whistlers are all generated at the equator, propagating along the magnetic field, and that the observed obliqueness is due to propagation effects. We also identified three types of electron distributions observed simultaneously with the whistler waves including beam-like, beam/flattop, and anisotropic distributions. The whistlers exhibited different characteristics depending on the observed electron distributions. The majority of the waveforms observed in our study have f/f_ce </- 0.5 and are observed primarily in the radiation belts simultaneously with anisotropic electron distributions. △ Less

Submitted 17 January, 2011; originally announced January 2011.

Comments: 18 pages, 5 figures, presented at 2010 Fall AGU