Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “DEEP LEARNING”

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.

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

At least 91 records · Page 5

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↗

Towards Scalable Deep Learning via I/O Analysis and Optimization

Deep learning systems have been growing in prominence as a way to automatically characterize objects, trends, and anomalies. Given the importance of deep learning systems, researchers have been investigating techniques to optimize such systems. An area of particular interest has been using large supercomputing systems to quickly generate effective deep learning networks: a phase often referred to as “training” of the deep learning neural network. As we scale existing deep learning frameworks—such as Caffe—on these large supercomputing systems, we notice that the parallelism can help improve the computation tremendously, leaving data I/O as the major bottleneck limiting the overall system scalability. In this paper, we first present a detailed analysis of the performance bottlenecks of Caffe on large supercomputing systems. Our analysis shows that the I/O subsystem of Caffe—LMDB—relies on memory-mapped I/O to access its database, which can be highly inefficient on large-scale systems because of its interaction with the process scheduling system and the network-based parallel filesystem. Based on this analysis, we then present LMDBIO, our optimized I/O plugin for Caffe that takes into account the data access pattern of Caffe in order to vastly improve I/O performance. Our experimental results show that LMDBIO can improve the overall execution time of Caffe by nearly 20-fold in some cases.

Pumma, Sarunya↗

Scalable Deep Learning via I/O Analysis and Optimization

Scalable deep neural network training has been gaining prominence because of the increasing importance of deep learning in a multitude of scientific and commercial domains. Consequently, a number of researchers have investigated techniques to optimize deep learning systems. Much of the prior work has focused on runtime and algorithmic enhancements to optimize the computation and communication. Despite these enhancements, however, deep learning systems still suffer from scalability limitations, particularly with respect to data I/O. This situation is especially true for training models where the computation can be effectively parallelized, leaving I/O as the major bottleneck. In fact, our analysis shows that I/O can take up to 90% of the total training time. Thus, in this article, we first analyze LMDB, the most widely used I/O subsystem of deep learning frameworks, to understand the causes of this I/O inefficiency. Based on our analysis, we propose LMDBIO—an optimized I/O plugin for scalable deep learning. LMDBIO includes six novel optimizations that together address the various shortcomings in existing I/O for deep learning. Lastly, our experimental results show that LMDBIO significantly outperforms LMDB in all cases and improves overall application performance by up to 65-fold on a 9,216-core system.

97 MATHEMATICS AND COMPUTING↗

S AP F LOWER : an automated tool for sap flow data preprocessing, gap-filling, and analysis using deep learning

Sap flow, a critical process in plant water use and ecosystem water cycles, is often measured using thermal dissipation probes (TDP) due to their ease of installation and continuous data collection. However, sap flow data frequently include noise, outliers, and gaps, creating challenges for analysis and requiring substantial manual processing. We developed S AP F LOWER , a tool that automates data preprocessing, model training, gap-filling, sapwood area scaling and modeling, and water use analysis. It integrates autocleaning, machine learning and deep learning models (e.g. random forest, Gaussian process regression, long short-term memory (LSTM), bidirectional LSTM (BiLSTM)), and efficient workflows to process sap flow data. S AP F LOWER can remove over 90% of noisy data while preserving legitimate variations and achieve high accuracy in gap-filling based on user-determined parameters. Random forest, LSTM, and BiLSTM models reduced root mean square error to 10% or less for long-term gaps. Model training and prediction can be performed efficiently within seconds. S AP F LOWER significantly enhances the efficiency and accessibility of TDP data analysis by automating complex tasks, enabling researchers without programming expertise to employ advanced techniques. Future improvements will focus on species-specific corrections for TDP and support for additional measurement methods. S AP F LOWER is openly available on GitHub (https://github.com/JiaxinWang123/SapFlower) and Zenodo (doi: 10.5281/zenodo.13665919).

ecosystem water balance↗

BUTTER - Empirical Deep Learning Dataset

The BUTTER Empirical Deep Learning Dataset represents an empirical study of the deep learning phenomena on dense fully connected networks, scanning across thirteen datasets, eight network shapes, fourteen depths, twenty-three network sizes (number of trainable parameters), four learning rates, six minibatch sizes, four levels of label noise, and fourteen levels of L1 and L2 regularization each. Multiple repetitions (typically 30, sometimes 10) of each combination of hyperparameters were preformed, and statistics including training and test loss (using a 80% / 20% shuffled train-test split) are recorded at the end of each training epoch. In total, this dataset covers 178 thousand distinct hyperparameter settings ("experiments"), 3.55 million individual training runs (an average of 20 repetitions of each experiments), and a total of 13.3 billion training epochs (three thousand epochs were covered by most runs). Accumulating this dataset consumed 5,448.4 CPU core-years, 17.8 GPU-years, and 111.2 node-years.

Array↗

Kinetic Deep Learning v0.1

Here, we present a method that uses protein levels to predict times series of metabolite concentrations. Understanding this type of pathway dynamics is important in order to predict the behavior of the pathway and, more pragmatically, to be able to design biological systems (such as strains bioengineered to produce chemical products) reliably. Typically, for this purpose, kinetic models consisting of differential equations based on the Michaelis-Menten dynamics have been used in the past. However, these methods can rarely produce good fits to measured data time series. Possibly, this happens because the kinetic constants are unknown or are different from the ones measured in vivo, or perhaps because Michaelis-Menten dynamics is not a satisfactory description. In order to improve the predictive nature of these kinetic models we have eliminated the Michaelis-Menten description of pathway dynamics and we have substituted it by algorithms that automatically learn these dynamics from previously obtained metabolomics and proteomics data using machine learning approaches. Specifically, kinetic deep learning uses deep learning to map proteomics time series to metabolite concentration time series, instead of learning the first metabolite derivative and integrating in (as in the first version of kinetic learning). This approach is shown to provide good to excellent results with a data set specifically collected for this purpose.

Garcia Martin, Hector [Joint BioEnergy Institute (↗

Fast uncertainty estimates in deep learning interatomic potentials

Deep learning has emerged as a promising paradigm to give access to highly accurate predictions of molecular and material properties. A common short-coming shared by current approaches, however, is that neural networks only give point estimates of their predictions and do not come with predictive uncertainties associated with these estimates. Existing uncertainty quantification efforts have primarily leveraged the standard deviation of predictions across an ensemble of independently trained neural networks. This incurs a large computational overhead in both training and prediction, resulting in order-of-magnitude more expensive predictions. Here, we propose a method to estimate the predictive uncertainty based on a single neural network without the need for an ensemble. This allows us to obtain uncertainty estimates with virtually no additional computational overhead over standard training and inference. We demonstrate that the quality of the uncertainty estimates matches those obtained from deep ensembles. We further examine the uncertainty estimates of our methods and deep ensembles across the configuration space of our test system and compare the uncertainties to the potential energy surface. Finally, we study the efficacy of the method in an active learning setting and find the results to match an ensemble-based strategy at order-of-magnitude reduced computational cost.

Chemistry↗

Evaluating On-Node GPU Interconnects for Deep Learning Workloads

Scaling deep learning workloads across multiple GPUs on a single node has become increasingly important in data analytics. A key question is how well a PCIe-based GPU interconnect can perform relative to a custom high-performance interconnect such as NVIDIA's NVLink. This paper evaluates two such on-node interconnects for eight NVIDIA Pascal P100 GPUs: (a) the NVIDIA DGX-1's NVLink 1.0 `hybrid cube mesh'; and (b) the Cirrascale GX8's two-level PCIe tree using dual SR3615 switch risers. To show the effects of a range of neural network workloads, we define a parameterized version of the popular ResNet. We define a workload intensity metric that characterizes the expected computation/communication ratio; we also locate AlexNet and GoogLeNet within that space. As expected, the DGX-1 typically has superior performance. However, when equalizing GPU SM frequencies, the GX8 is very competitive on all ResNet workloads. With 8 GPUs, the GX8 can outperform the DGX-1 on all-to-all reductions by 10% for medium-sized payloads; and in rare cases, the GX8 slightly outperforms on ResNet.

Tallent, Nathan R.↗

Do optimization methods in deep learning applications matter?

With advances in deep learning, exponential data growth and increasing model complexity, developing efficient optimization methods are attracting much research attention. Several implementations favor the use of Conjugate Gradient (CG) and Stochastic Gradient Descent (SGD) as being practical and elegant solutions to achieve quick convergence, however, these optimization processes also present many limitations in learning across deep learning applications. Recent research is exploring higher-order optimization functions as better approaches, but these present very complex computational challenges for practical use. Comparing first and higher-order optimization functions, in this paper, our experiments reveal that Levemberg-Marquardt (LM) significantly supersedes optimal convergence but suffers from very large processing time increasing the training complexity of both, classification and reinforcement learning problems. Our experiments compare off-the-shelf optimization functions(CG, SGD, LM and L-BFGS) in standard CIFAR, MNIST, CartPole and FlappyBird experiments.The paper presents arguments on which optimization functions to use and further, which functions would benefit from parallelization efforts to improve pretraining time and learning rate convergence.

Ozyildirim, Buse Melis↗

Recent advances and applications of deep learning methods in materials science

Deep learning (DL) is one of the fastest-growing topics in materials data science, with rapidly emerging applications spanning atomistic, image-based, spectral, and textual data modalities. DL allows analysis of unstructured data and automated identification of features. The recent development of large materials databases has fueled the application of DL methods in atomistic prediction in particular. In contrast, advances in image and spectral data have largely leveraged synthetic data enabled by high-quality forward models as well as by generative unsupervised DL methods. In this article, we present a high-level overview of deep learning methods followed by a detailed discussion of recent developments of deep learning in atomistic simulation, materials imaging, spectral analysis, and natural language processing. For each modality we discuss applications involving both theoretical and experimental data, typical modeling approaches with their strengths and limitations, and relevant publicly available software and datasets. We conclude the review with a discussion of recent cross-cutting work related to uncertainty quantification in this field and a brief perspective on limitations, challenges, and potential growth areas for DL methods in materials science.

36 MATERIALS SCIENCE↗

Embracing the unknown: Why uncertainty matters in deep learning

Most cutting-edge deep learning models are focused on only one thing: improving prediction accuracy. But in the real world, especially in high-stakes domains like science, medicine, and safety (e.g. self-driving cars), knowing what your model doesn't know can be just as important. My goal for this talk is to convince you that uncertainty quantification (UQ) is essential in deep learning. We’ll explore how existing UQ methods work, what they still get wrong, and how ideas from disciplines like astrophysics and environmental science are key to driving the development of better (more trustworthy) deep learning methods.

Nevin, Becky [Fermilab]↗

The use of digital thread for reconstruction of local fiber orientation in a compression molded pin bracket via deep learning

A deep convolutional neural network (DCNN) was used for microstructure reconstruction using artificial intelligence (MR-AI) by predicting local average fiber orientation distributions (FOD) in a 3D prepreg platelet molded composite (PPMC) pin bracket. To train the MR-AI model, surface strain fields from residual stresses simulated in PPMC plates were used as the input to the DCNN. A training dataset included PPMC plates with various degrees of global fiber alignment, based on the information obtained from high-fidelity flow simulation of a pin bracket. Further, the MR-AI model was then deployed to analyze FOD in the 3D pin bracket by conducting thermo-elastic residual stress analysis. Initially, the MR-AI model was established entirely on the synthetic simulation data. Then, a μCT scan of a physically molded pin bracket was used to create a finite element model that provided data for additional validation of the DCNN model. For the μCT scan finite element pin bracket the MR-AI model predicted the distribution of fiber orientation tensor components with MAE of 0.10 indicating a global prediction error of 10%. For the flow simulated pin bracket, the MR-AI model predicted the distribution of fiber orientation tensor components with a global prediction error of 11%. The MR-AI model showed the ability to predict regions of varying alignment in the base and flange of the pin bracket. The proposed MR-AI methodology allows for rapid prediction of FOD in geometrically complex parts and offers a promising path to detecting unique fiber orientation states in molded components.

42 ENGINEERING↗

Identifying the nature of the QCD transition in relativistic collision of heavy nuclei with deep learning

Using deep convolutional neural network (CNN), the nature of the QCD transition can be identified from the final-state pion spectra from hybrid model simulations of heavy-ion collisions that combines a viscous hydrodynamic model with a hadronic cascade “after-burner”. Two different types of equations of state (EoS) of the medium are used in the hydrodynamic evolution. The resulting spectra in transverse momentum and azimuthal angle are used as the input data to train the neural network to distinguish different EoS. Different scenarios for the input data are studied and compared in a systematic way. A clear hierarchy is observed in the prediction accuracy when using the event-by-event, cascade-coarse-grained and event-fine-averaged spectra as input for the network, which are about 80%, 90% and 99%, respectively. A comparison with the prediction performance by deep neural network (DNN) with only the normalized pion transverse momentum spectra is also made. High-level features of pion spectra captured by a carefully-trained neural network were found to be able to distinguish the nature of the QCD transition even in a simulation scenario which is close to the experiments.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Data for The utility of transfer learning to improve the performance of deep learning in axon segmentation

The utility of transfer learning to improve the performance of deep learning in axon segmentation Data Data: All the input and labeled volumes tf-logs: Tensorflow logs, view with command "tensorboard --logdir [name of folder]" Model Weights: model_weights: the argument list under variable combo indicate 1) no oversampling, 2) no rotation, 3) no learn scheduler, and 4) flipping on all three dimensions, and the additional values indicate 5) elastic deformation percentage, 6) rotate deformation percentage, 7) layer setting , 8) learning rate, and 9) training/validation/test data division suffix (leave '' if not using suffix). Results: Output from inference segment_total_results_validation_final: All validation results and calculations segment_total_results: All test results and calculations Authors The modified code was created for a paper by: Marjolein Oostrom, Michael A. Muniak, Rogene Eichler West, Sarah Akers, Paritosh Pande, Moses Obiri, Wei Wang, Kasey Bowyer, Zhuhao Wu, Lisa Bramer, Tianyi Mao, Bobbie Jo Webb-Robertson The work is adapted from Github TrailMap, which was created by Albert Pun and Drew Friedmann Acknowledgments MO, RMEW, SA, MO, LB, BJWR were supported by the Laboratory Directed Research and Development at Pacific Northwest National Laboratory (PNNL), a Department of Energy facility operated by Battelle under contract DE-AC05-76RLO01830. WW, KB, and ZW were supported in part by a NIH/BRAIN Initiative Grant RF1MH128969. MAM and TM were supported by two NIH/BRAIN Initiative Grants R01NS104944, RF1MH120119 and NIH R01NS081071. This research is affiliated with the Pacific northwest bioMedical Innovation Co-laboratory (PMedIC) collaboration between OHSU and PNNL.

Oostrom, Marjolein T↗