Search NASA⌕ Search

Engineering topics

Dumont, Vincent

Publications and source records attributed to Dumont, Vincent.

Engagement: Hyperparameter Optimization of Generative Adversarial Network Models for High-Energy Physics Simulations

We present our SciDAC FASTMath-HEP partnership results for tuning generative adversarial models (GANs) for high energy physics applications. The GANs are used in hybrid simulations to accelerate otherwise time-consuming computations. We optimize for both, prediction accuracy and variability with the goal to find GAN architectures that are reliable and robust.

high energy physics↗

Surrogate Model Based Optimization for Finding Robust Deep Learning Model Architectures

Deep Learning (DL) models are increasingly used throughout the sciences. However, their performance and usefulness depend greatly on their architecture which is defined by hyperparameters such as the number of nodes, layers, the learning rate, etc. Tuning these hyperparameters is time-consuming because evaluating their performance requires a lengthy training step. Stochastic optimizers used in training lead to performance variability and potentially prediction reliability issues. In this talk, we will describe an automated optimization method based on surrogate models and active learning strategies for tuning DL model architectures. We take into account the prediction variability with the goal to identify architectures that make reliable and robust predictions. We demonstrate our developments on an application arising in particle physics.

deep learning↗

HYPPO: A Surrogate-Based Multi-Level Parallelism Tool for Hyperparameter Optimization

We present a new software, HYPPO, that enables the automatic tuning of hyperparameters of various deep learning (DL) models. Unlike other hyperparameter optimization (HPO) methods, HYPPO uses adaptive surrogate models and directly accounts for uncertainty in model predictions to find accurate and reliable models that make robust predictions. Using asynchronous nested parallelism, we are able to significantly alleviate the computational burden of training complex architectures and quantifying the uncertainty. HYPPO is implemented in Python and can be used with both TensorFlow and PyTorch libraries. We demonstrate various software features on time-series prediction and image classification problems as well as a scientific application in computed tomography image reconstruction. Finally, we show that (1) we can reduce by an order of magnitude the number of evaluations necessary to find the most optimal region in the hyperparameter space and (2) we can reduce by two orders of magnitude the throughput for such HPO process to complete.

adaptation models↗

Search for topological defect dark matter with a global network of optical magnetometers

Ultralight bosons such as axion-like particles are viable candidates for dark matter. They can form stable, macroscopic field configurations in the form of topological defects that could concentrate the dark matter density into many distinct, compact spatial regions that are small compared with the Galaxy but much larger than the Earth. Here we report the results of the search for transient signals from the domain walls of axion-like particles by using the global network of optical magnetometers for exotic (GNOME) physics searches. We search the data, consisting of correlated measurements from optical atomic magnetometers located in laboratories all over the world, for patterns of signals propagating through the network consistent with domain walls. The analysis of these data from a continuous month-long operation of GNOME finds no statistically significant signals, thus placing experimental constraints on such dark matter scenarios.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

HYPPO v1.0.0

HYPPO is a hyperparameter optimization (HPO) software that uses surrogate modeling and uncertainty quantification to provide reliable neural network architectures. Similar state-of-the-art HPO technologies do not use such features and mostly perform random searches to explore the parameter space, which can easily become computationally expensive. Using HYPPO, the user is able to reduce by an order of magnitude the number of evaluations necessary to identify the most optimal region in the hyperparameter space. Finally, using asynchronous nested parallelism, we are able to decrease by two orders of magnitude the amount of time required to complete the HPO process.

Dumont, Vincent↗

Machine Learning for Distributed Acoustic Sensing data (MLDAS) v1.0.1

MLDAS is a Python-written package for exploratory data analysis and deep learning training on Distributed Acoustic Sensing data. The machine learning tools are powered by the PyTorch library and designed to work efficiently on large scale datasets using parallel computing. Various SLURM scripts as well as a tutorial have also been made available to allow geophysicists to quickly and easily implement the available tools in their analysis workflow on supercomputer facilities.

Dumont, Vincent↗

Four direct measurements of the fine-structure constant 13 billion years ago

Observations of the redshift z = 7.085 quasar J1120+0641 are used to search for variations of the fine structure constant, a, over the redshift range 5:5 to 7:1. Observations at z = 7:1 probe the physics of the universe at only 0.8 billion years old. These are the most distant direct measurements of a to date and the first measurements using a near-IR spectrograph. A new AI analysis method is employed. Four measurements from the x - shooter spectrograph on the Very Large Telescope (VLT) constrain changes in a relative to the terrestrial value (α 0 ). The weighted mean electromagnetic force in this location in the universe deviates from the terrestrial value by Δα/α = (α z - α 0 )/α 0 = (-2:18 ± 7:27) × 10 -5 , consistent with no temporal change. Combining these measurements with existing data, we find a spatial variation is preferred over a no-variation model at the 3:9σ level.

79 ASTRONOMY AND ASTROPHYSICS↗

Septin 7 reduces nonmuscle myosin IIA activity in the SNAP23 complex and hinders GLUT4 storage vesicle docking and fusion

Glomerular epithelial cells, podocytes, are insulin responsive and can develop insulin resistance. Here, we demonstrate that the small GTPase septin 7 forms a complex with nonmuscle myosin heavy chain IIA (NMHC-IIA; encoded by MYH9), a component of the nonmuscle myosin IIA (NM-IIA) hexameric complex. We observed that knockdown of NMHC-IIA decreases insulin-stimulated glucose uptake into podocytes. Both septin 7 and NM-IIA associate with SNAP23, a SNARE protein involved in GLUT4 storage vesicle (GSV) docking and fusion with the plasma membrane. We observed that insulin decreases the level of septin 7 and increases the activity of NM-IIA in the SNAP23 complex, as visualized by increased phosphorylation of myosin regulatory light chain. Also knockdown of septin 7 increases the activity of NM-IIA in the complex. The activity of NM-IIA is increased in diabetic rat glomeruli and cultured human podocytes exposed to macroalbuminuric sera from patients with type 1 diabetes. Collectively, the data suggest that the activity of NM-IIA in the SNAP23 complex plays a key role in insulin-stimulated glucose uptake into podocytes. Furthermore, we observed that septin 7 reduces the activity of NM-IIA in the SNAP23 complex and thereby hinders GSV docking and fusion with the plasma membrane. - Highlights: • Septin 7, nonmuscle myosin heavy chain IIA (NMHC-IIA) and SNAP23 form a complex. • Knockdown of septin 7 increases NM-IIA activity in the SNAP23 complex. • Insulin decreases septin 7 level and increases NM-IIA activity in the SNAP23 complex. • Septin 7 hinders GSV docking/fusion by reducing NM-IIA activity in the SNAP23 complex.

60 APPLIED LIFE SCIENCES↗