SEARCH · Search NASA

Results for “PyTorch”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4

All sources NASA NTRS DOE OSTI

DOE OSTI2024

Impacts of floating-point non-associativity on reproducibility for HPC and deep learning applications

Run to run variability in parallel programs caused by floating-point non-associativity has been known to significantly affect reproducibility in iterative algorithms, due to accumulating errors. Non-reproducibility can critically affect the efficiency and effectiveness of correctness testing for stochastic programs. Recently, the sensitivity of deep learning training and inference pipelines to floating-point non-associativity has been found to sometimes be extreme. It can prevent certification for commercial applications, accurate assessment of robustness and sensitivity, and bug detection. New approaches in scientific computing applications have coupled deep learning models with high-performance computing, leading to an aggravation of debugging and testing challenges. Here we perform an investigation of the statistical properties of floating-point non-associativity within modern parallel programming models, and analyze performance and productivity impacts of replacing atomic operations with deterministic alternatives on GPUs. We examine the recently-added deterministic options in PyTorch within the context of GPU deployment for deep learning, uncovering and quantifying the impacts of input parameters triggering run to run variability and reporting on the reliability and completeness of the documentation. Finally, we evaluate the strategy of exploiting automatic determinism that could be provided by deterministic hardware, using the Groq LPUTM accelerator for inference portions of the deep learning pipeline. We demonstrate the benefits that a hardware-based strategy can provide within reproducibility and correctness efforts.

Shanmugavelu, Sanjif↗

DOE OSTI2024

Projective Integral Updates for High-Dimensional Variational Inference

Variational inference is an approximation framework for Bayesian inference that seeks to improve quantified uncertainty in predictions by optimizing a simplified distribution over parameters to stand in for the full posterior. Capturing model variations that remain consistent with training data enables more robust predictions by reducing parameter sensitivity. This work introduces a fixed-point optimization for variational inference that is applicable when every feasible log density can be expressed as a linear combination of functions from a given basis. In such cases, the optimizer becomes a fixed-point of projective integral updates. When the basis spans univariate quadratics in each parameter, the feasible distributions are Gaussian mean-fields and the projective integral updates yield quasi-Newton variational Bayes (QNVB). Other bases and updates are also possible. Since these updates require high-dimensional integration, this work begins by proposing an efficient quasirandom sequence of quadratures for mean-field distributions. Each iterate of the sequence contains two evaluation points that combine to correctly integrate all univariate quadratic functions and, if the mean-field factors are symmetric, all univariate cubics. More importantly, averaging results over short subsequences achieves periodic exactness on a much larger space of multivariate polynomials of quadratic total degree. The corresponding variational updates require four loss evaluations with standard (not second-order) backpropagation to eliminate error terms from over half of all multivariate quadratic basis functions. Furthermore, this integration technique is motivated by first proposing stochastic blocked mean-field quadratures, which may be useful in other contexts. A PyTorch implementation of QNVB allows for better control over model uncertainty during training than competing methods. Experiments demonstrate superior generalizability for multiple learning problems and architectures.

Gaussian mean-field↗

DOE OSTI2022

Punzi-loss: a non-differentiable metric approximation for sensitivity optimisation in the search for new particles

We present the novel implementation of a non-differentiable metric approximation and a corresponding loss-scheduling aimed at the search for new particles of unknown mass in high energy physics experiments. We call the loss-scheduling, based on the minimisation of a figure-of-merit related function typical of particle physics, a Punzi-loss function, and the neural network that utilises this loss function a Punzi-net. We show that the Punzi-net outperforms standard multivariate analysis techniques and generalises well to mass hypotheses for which it was not trained. This is achieved by training a single classifier that provides a coherent and optimal classification of all signal hypotheses over the whole search space. Our result constitutes a complementary approach to fully differentiable analyses in particle physics. We implemented this work using PyTorch and provide users full access to a public repository containing all the codes and a training example.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

DOE OSTI2025

TorchBraid: High-Performance Layer-Parallel Training of Deep Neural Networks with MPI and GPU Acceleration

TorchBraid is a high-performance implementation of layer-parallel training for deep neural networks (DNNs) supporting MPI-based parallelism and GPU acceleration. Layer-parallel training has been developed to overcome the serialization inherent in forward and backward propagation of DNNs that limits utilization of computational resources in the strong scaling limit. To achieve this, TorchBraid integrates the PyTorch neural network framework with the state-of-the-art XBraid time-parallel library. Furthermore, this article presents the use and performance of TorchBraid, in addition to solutions for overcoming the algorithmic challenges inherent in combining automatic differentiation with layer-parallel. Results are presented with and without GPU acceleration for the Tiny ImageNet and MNIST image classification data sets, as well as recurrent neural networks. Overall, TorchBraid enables fast training of DNNs, both in a strong and weak scaling context. In addition to the TorchBraid software, several new advances in applying layer-parallel algorithms are detailed. Integration of layer-parallel with data-parallel algorithms is presented for the first time, showing the computational advantages of the combination. Standard deep learning techniques, like batch-normalization, are developed for layer-parallel training. Finally, a new approach combining layer-parallel with spatial coarsening in order to accelerate training for 3D image classification shows roughly a 10× speedup over serial execution.

Layer-parallel↗

DOE OSTI2026

The Memory Scaling of Reverse-Mode Differentiation in Particle Accelerator Simulations with Space Charge

The recent development of differentiable simulation codes for particle accelerators has enabled gradient-based workflows that promise finer control and more realistic modeling of accelerator facilities. However, when using reverse-mode automatic differentiation, the memory usage continuously increases during the simulation, and can potentially exceed the available hardware memory - especially when costly space charge computation is included. To study the memory requirements for differentiable simulations, we have implemented space charge in Cheetah, a PyTorch-based beam tracking code that supports reverse-mode differentiation. We find that the memory usage for reverse-mode differentiation grows linearly with the number of macroparticles and cells, and that it is proportional to the number of space charge kicks involved in the simulation. This general scaling can be used to evaluate whether a given differentiable simulation is feasible given hardware memory constraints.

Dhamrait, Arjun↗

DOE OSTI2019

Remote Sensor Design for Visual Recognition with Convolutional Neural Networks

While deep learning technologies for computer vision have developed rapidly since 2012, modeling of remote sensing systems has remained focused around human vision. In particular, remote sensing systems are usually constructed to optimize sensing cost-quality trade-offs with respect to human image interpretability. While some recent studies have explored remote sensing system design as a function of simple computer vision algorithm performance, there has been little work relating this design to the state-of-the-art in computer vision: deep learning with convolutional neural networks. We develop experimental systems to conduct this analysis, showing results with modern deep learning algorithms and recent overhead image data. Our results are compared to standard image quality measurements based on human visual perception, and we conclude not only that machine and human interpretability differ significantly, but that computer vision performance is largely self-consistent across a range of disparate conditions. This repository contains all the code required to replicate the results of our research paper, and the corresponding docker environment in which that code can be executed. All code is written in Python3 and utilizes the PyTorch library (v1.0) for neural network model training and evaluation. Jupyter notebooks are used to visualize images transformed with our code, in addition to plotting experimental results. We also include the specific parameter files used to execute our experiments, with the intent that these experiments can be replicated.

Zelinski, michaelE.↗

DOE OSTI2019

Efficent XAI Information Saliency Tool

This software is an implementation which is described in: "Efficient Saliency Maps for Explainable AI" (https://openreview.net/forum?id=ryxf9CEKDr). The toolkit is located in: https://mybitbucket.llnl.gov/projects/SAL/repos/smoe_lovi/browse. It is a set of of PyTorch related python scripts for extracting the what parts of an image are most salient to a deep neural network in an efficient manner. This source will allow one to replicate the results in the paper referenced above. Is there

MUNDHENK, TERRELLN↗

DOE OSTI2019

STEMDL

The code "STEMDL" provides highly-optimized multi-gpu implementations of DL models of applicability in electron microscopy via the use of mixed-precision arithmetic, native graph construction and evaluation in Tensorflow, and adaptive scaling of learning parameters to avoid numerical instabilities due to reduced precision. STEMDL also uses low-level methods from the Horovod library for gradient communication and reduction and adds an additional optimization by allowing the user to scheduling the gradient reduction operations based on their position within the layers of a neural net. STEMDL provides optimized I/O routines for data staging and data reading/writing in both Tensorflow and Pytorch using multi-processing and asynchronous transfer between host and device. .

Laanait, Nouamane↗

DOE OSTI2020

TorchBraid

Torchbraid is a python package designed for interaction with the PyTorch library. It enables training of neural networks in a layer-parallel fashion.SAND2020-12228 M Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

Cyr, Eric↗

DOE OSTI2020

Concurrent Relaxation through Accelerated Deep Learning

CRADL captures performance metrics of machine learning algorithms operating on mesh data from multiphysics codes This proxy application is a tool to explore scalability of inference on HPC platforms, and also gather performance metrics for inference on new machine learning specific hardware. CRADL is designed to give users as fine a control as possible over an inference simulation. Users may select the number of cycles, amount of data, and batch size to pass to the accelerator of choice. Additionally the user may select a number of performance optimization libraries and flags. CRADL comes packaged with a repository of anonymized multi-physics simulation data, as well as a pretrained model for inference. The code allows a user to load their own pre-trained model and data if they wish. The code can operate in multiple parallelization schemes, with performance enhancing options such as half-precision libraries, PyTorch benchmarking, and pinned memory with non-blocking data transfers.

Zieb, KristoferJ.↗

DOE OSTI2020

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↗

DOE OSTI2020

Euclidean neural networks (e3nn) v1.0

E(3) is the Euclidean group in dimension 3. That is the group of rotations, translations and mirror. e3nn is a pytorch library that aims to create E(3) equivariant neural networks.

Lapchevskyi, Kostiantyn↗

DOE OSTI2021

FlexDRL v1.0

FlexDRL is a framework that can be used as a benchmarking tool to evaluate different Deep Reinforcement Learning algorithms in a building energy management setting. OpenAI Gym was used as wrapper to the co-simulation environment (EnergyPlus for envelope and HVAC and Modelica for battery and PV) to interface with DRL algorithms, which are developed in PyTorch. In order to facilitate the usage of FlexDRL, all the required dependencies have been packaged into a Docker container.

Touzani, Samir↗

DOE OSTI2021

Graphene U-Net v1

Graphene U-Net is a library that provides a simplified platform for training deep neural networks for the task of microscopy image segmentation. It contains functions and classes that make the process of training a neural network simple such that non-ML experts can train and evaluate models on their own datasets. It uses Pytorch as a backend and can run on both CUDA-enable GPUs and CPUs. It contains the main library file Microscopy_Unet.py as well as unet.py which contains the UNET model used for this software. This can be replaced with any other fully convolutional deep learning architecture with relative ease.

Sadre, Robbie↗

DOE OSTI2021

stacked predictive sparse decomposition (SPSD) v1.0

It is an unsupervised representation algorithm used to learn morphometric properties from cellular objects. Many similar/identical open source implementations are widely available in tensorflow, pytorch, etc. And ours is implemented in matlab.

Chang, Hang↗

DOE OSTI2021

pyMSDtorch v0.1.0

pyMSDtorch is a python implementation of the MSDNet developed by Sethian and Pelt [1] for image analyses, using the pytorch framework. The availability these networks in mainstream ML packages makes it easier to attach standard expansions and other approaches to make it more flexible and user friendly.

Zwart, PetrusH↗

DOE OSTI2021

HydraGNN

Distributed PyTorch implementation of multi-headed graph convolutional neural networks

Lupo Pasini, Massimiliano [Oak Ridge National Lab.↗

DOE OSTI2021

spdlayers

Symmetric Positive Definite (SPD) enforcement layers for PyTorch. Regardless of the input, the output of these layers will always be a SPD tensor! The `Cholesky` layer uses a cholesky factorization to enforce SPD, and the `Eigen` layer uses an eigendecomposition to enforce SPD. Both layers take in some tensor of shape `[batch_size, input_shape]` and output a SPD tensor of shape`[batch_size, output_shape, output_shape]`. The relationship between input and output is defined by the following. ```python input_shape = sum([i for i in range(output_shape + 1)]) ``` The layers have no learnable parameters, and merely serve to transform a vector space to a SPD matrix space.

Jekel, CharlesF.↗