Search NASA⌕ Search

SEARCH · Search NASA

Results for “model performance”

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 523 records · Page 29
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Powered By SAM [Slides]

The System Advisor Model(TM) (SAM) is a free, open-source desktop application for techno-economic analysis of energy technologies. By combining detailed performance modeling with financial analysis, SAM allows users to assess technology trade-offs, explore future scenarios, and make informed decisions about energy investments. Users also have access to model details and the ability to embed SAM's core models in their own applications. This webinar, hosted by National Laboratory of the Rockies researchers Janine Keith and Matt Prilliman, highlights how this widely used modeling tool supports data-driven decision-making for energy systems.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
DOE OSTI2021

A Review of Machine Learning Classification Using Quantum Annealing for Real-World Applications

Optimizing the training of a machine learning pipeline helps in reducing training costs and improving model performance. One such optimizing strategy is quantum annealing, which is an emerging computing paradigm that has shown potential in optimizing the training of a machine learning model. The implementation of a physical quantum annealer has been realized by D-wave systems and is available to the research community for experiments. Recent experimental results on a variety of machine learning applications using quantum annealing have shown interesting results where the performance of classical machine learning techniques is limited by limited training data and high dimensional features. This article explores the application of D-wave’s quantum annealer for optimizing machine learning pipelines for real-world classification problems. Finally, we review the application domains on which a physical quantum annealer has been used to train machine learning classifiers. We discuss and analyze the experiments performed on the D-Wave quantum annealer for applications such as image recognition, remote sensing imagery, computational biology, and particle physics. We discuss the possible advantages and the problems for which quantum annealing is likely to be advantageous over classical computation.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Performance Evaluation of a Two-Dimensional Flood Model on Heterogeneous High-Performance Computing Architectures

This paper describes the implementation of a two-dimensional hydrodynamic flood model with two different numerical schemes on heterogeneous high-performance computing architectures. Both schemes were able to solve the nonlinear hyperbolic shallow water equations using an explicit upwind first-order approach on finite differences and finite volumes, respectively, and were conducted using MPI and CUDA. Four different test cases were simulated on the Summit supercomputer at Oak Ridge National Laboratory. Both numerical schemes scaled up to 128 nodes (768 GPUs) with a maximum 98.2x speedup of over 1 GPU. The lowest run time for the 10 day Hurricane Harvey event simulation at 5 meter resolution (272 million grid cells) was 50 minutes. GPUDirect communication proved to be more convenient than the standard communication strategy. Both strong and weak scaling are shown.

Sharif, Md Bulbul↗
DOE OSTI2024

Quantum Annealing for Real-World Machine Learning Applications

Optimizing the training of a machine learning pipeline is important for reducing training costs and improving model performance. One such optimizing strategy is quantum annealing, which is an emerging computing paradigm that has shown potential in optimizing the training of a machine learning model. The implementation of a physical quantum annealer has been realized by D-Wave systems and is available to the research community for experiments. Recent experimental results on a variety of machine learning applications have shown interesting results especially under the conditions where the performance of classical machine learning techniques are limited such as limited training data and high dimensional features. This chapter explores the application of D-Wave’s quantum annealer for optimizing machine learning pipelines for real-world classification problems. We review the application domains on which a physical quantum annealer has been used to train machine learning classifiers. We discuss and analyze the experiments performed on the D-Wave quantum annealer for applications such as image recognition, remote sensing imagery, security, computational biology, biomedical sciences, and physics. We discuss the possible advantages and the problems for which quantum annealing is likely to be advantageous over classical computation.

Kumar nath, Rajdeep↗
DOE OSTI2025

Benchmarking Near-Surface Winds in the HRRR Analyses Using Multisource Observations over Complex Terrain in the Southeastern United States

Wind energy plays a crucial role in sustainable power generation, yet its full potential in the southeast United States (SEUS) remains underexplored. This study advances wind resource assessment in the SEUS using existing regional modeling and multisource observations. We find that terrain complexity, influenced by orography and forest canopy, significantly impacts the accuracy of modeled wind speed. While the High-Resolution Rapid Refresh (HRRR) model effectively simulates wind profiles over flat, nonforested terrain, larger errors are produced above forest canopies, and particularly for those stands in hilly and mountainous terrain where there are no observations aloft that could be used by HRRR’s data assimilation software. Seasonal variations, e.g., changes in the leaf area index, further complicate the relationship between terrain complexity and simulation errors. Here, our findings emphasize the critical need for comprehensive wind profile measurements, extending from below the canopy to height above the canopy and through the lower planetary boundary layer, to fully quantify model performance over and near forests, and for improving wind resource assessment, planning, and management. This study provides valuable insights not only for further model development but also for future field campaign deployment with the goal of further improving our understanding of wind resources in the region.

Complex terrain↗
DOE OSTI2023

Evaluating Production Implications of Pressure Maintenance in Unconventional Oil and Gas Wells using a Machine Learning Modeling Approach: Case Study in the Permian Basin

This study implements the proprietary machine learning-based model (model) developed under the 2022 study titled “Evaluating the Impact of Proprietary Oil & Gas Data on Machine Learning Model Performance Using a Quasi-Experimental Analytical Approach” for forecasting unconventional oil and gas production using well data from the Permian Basin. The model, developed using an exclusive dataset that includes time series production data from an operator in the Permian Basin, is designed to jointly predict daily oil, gas, and water production for horizontal wells as a function of bottomhole pressure drawdown, spatial placement across the study domain, and well-completion attributes. In this study, the model was explicitly applied to explore its utility to evaluate the impact of varying drawdown strategies on the production forecast of a well in the Permian Basin dataset. Managing pressure drawdown has been identified as a way to improve estimated ultimate recovery (EUR) from unconventional shale wells due to the stress-dependent nature of fractures in shale reservoirs. Research has shown that applying a lower pressure drawdown helps to maintain the reservoir conductivity, resulting in higher productivity over the life of a well. Historic bottomhole pressure data from the well over time was used as a benchmark from which to set more and less aggressive pressure decline rates as bounding modeling cases. All pressure decline rates/strategies were forecasted over 5 years, and the model was used to generate oil, water, and gas prediction over the same timeframe. This report presents the results from the production forecast and discusses potential operational and economic implications, with contrasting perspectives between well productivity and profitability given typical oil and gas economics and the volatility in the oil and gas market.

02 PETROLEUM↗
DOE OSTI2022

Decoding Golden Eagle Movement Behavior from High-Resolution, Variable-Rate Telemetry Data Through Bayesian Filtering

The recent advances in animal tracking technology have enabled the collection of a vast amount of in situ data regarding the movement of wildlife at high spatiotemporal resolution. These data are usually available at variable time resolutions and contains noise (error) originating from GPS fixes. Decoding movement characteristics, particularly of flying animals, from telemetry data while handling these factors is a challenging yet important task for conservation purposes. Typically, this task is broken into two subtasks: resampling, and model calibration. The resampling subtask converts the variable rate positional data into a constant time interval data, while the model calibration subtask uses the resampled data to tune time-invariant parameters of the proposed models. For telemetry data at high temporal resolutions (order of 1 second), it is very challenging to decouple noise from actual movements using interpolation-based resampling techniques. Any errors introduced during resampling can significantly alter the the calibration and prediction attributes of the movement model. We address this problem through a unified Bayesian state-space framework that can handle both the resampling and calibration tasks in a single step. In addition, we use the speed and heading of the bird from telemetry data to regularize the position information of the bird. We use a Kalman filtering approach to include these nonlinearly related motion parameters within the state space framework. We cross-validated to quantify how this inclusion affects the model performance in estimating true bird movements. The relationship between the true state of the bird and environmental and topographical covariates is then represented parametrically. These parameters are then tuned using stochastic sampling strategies like Markov Chain Monte Carlo (MCMC). We use the telemetry data collected from golden eagles in the western USA to demonstrate the applicability of this approach to build a predictive, probabilistic movement model. Our preliminary results show that this approach provides improved predictive performance in terms of capturing higher-order motion parameters such as angular and horizontal accelerations, which may have simpler and more direct relationships with environmental covariates than corresponding speeds. In this talk, we will demonstrate how this state-space approach benefits the prediction capabilities of a movement model in simulating golden eagle paths through a wind power plant in Wyoming given certain atmospheric conditions. The model outcomes are aimed at informing mitigation strategies that can minimize the potential for collisions of golden eagles with wind turbines.

Bayesian methods↗
DOE OSTI2024

Comparative Assessment of U-Net-Based Deep Learning Models for Segmenting Microfractures and Pore Spaces in Digital Rocks

Segmentation of high-resolution X-ray microcomputed tomography (µCT) images is crucial in digital rock physics (DRP), affecting the characterization and analysis of microscale phenomena in the porous media. The complexity of geological structures and nonideal scanning conditions pose significant challenges to conventional image segmentation approaches. Motivated by the recent increasing popularity of deep learning (DL) techniques in image processing, this work undertakes a comparative study of DL models, specifically U-Net and its variants, for segmenting multiple targets with distinguished features in digital rocks, including discrete fracture networks (DFNs), pore spaces, and solid rock. Particularly, DFNs have a smaller volumetric fraction over others, bringing in a substantial challenge of imbalanced segmentation. The primary focus is to evaluate the architecture and feature enhancement strategies of various DL models, including U-Net, attention U-Net, residual U-Net, U-Net++, and residual U-Net++. The models were designed as 2.5D, utilizing a central 2D image and its two adjacent upper and lower 2D images as input to provide a pseudo-3D context. In addition, because the ground truth of segmentation was unknown for real-world digital rocks, we created a benchmark data set following the inverse operations of segmentation. The data synthesis started from the label images (i.e., solid rock, pore spaces, and DFNs), followed by simulating partial volume blurring, adding random background noise, and introducing ring artifacts to mimic real raw X-ray µCT images. The data set, which included various rock types (i.e., sandstone and artificial data), scanning resolution, and magnitudes of noise and artifacts, was divided into training and testing data sets with a 90% and 10% ratio, respectively. Moreover, in addition to the conventional pixel-wise evaluation metrics, the physics-based metric of the lattice-Boltzmann method (LBM) simulated permeability provided more comprehensive assessments. The results demonstrated that the residual connections, nested architectures, and redesigned skip connections contribute to the model performance and give the residual U-Net++ the highest accuracy. The improvements were mainly on the boundaries and small targets, especially the DFNs, which dominate the interconnectivity and therefore affect the permeability greatly. This study also rigorously evaluated the efficiency and generalization of each model, demonstrating that the sophisticated architectures achieved excellent practicability and maintained robust performance on completely unseen data, ensuring their suitability for diverse and challenging DRP applications.

58 GEOSCIENCES↗
DOE OSTI2021

New insights into fuel blending effects: Intermolecular chemical kinetic interactions affecting autoignition times and intermediate-temperature heat release

Fuel blending effects on chemically-dominated fuel properties, such as gasoline anti-knock quality, are influenced by fundamental chemical kinetic interactions between the blending agent and the base fuel. Historically, quantification of such interactions has focused on changes to the radical pool, including $\dot{O}$H and HO 2 , while intermolecular interactions pertaining to carbonated, non-fuel-specific intermediates are typically overlooked. In this regard, this work aims to derive new insight into intermolecular chemical kinetic interactions that are intrinsic to fuel blending effects via a case study on blends of 0–30% ethanol (by volume) into FGF-LLNL (a multi-component gasoline surrogate for FACE-F research gasoline) using a rapid compression machine at a diluted/stoichiometric fuel loading, compressed pressure of 40 bar and low- to intermediate-temperature regimes that are representative of boosted SI engine operation. Ethanol blending effects on the intermediate temperature heat release (ITHR) of FGF-LLNL are characterized using experimental measurements, where ethanol is found to promote the extent of ITHR and suppress the transition from ITHR to main ignition. Chemical kinetic modeling is undertaken using a recently updated gasoline surrogate model. Sensitivity analyses on ITHR characteristics further corroborate the ethanol blending effects, and highlight the significant dependence of ITHR on both fuel-specific and non-fuel-specific reactions. An approach allowing comprehensive characterization of the complex intermolecular chemical kinetic interactions between constitutes in a fuel blend is then proposed. Application of the approach to FGF-LLNL/E0–E30 reveals that ethanol perturbs the heat release and autoignition characteristics of FGF-LLNL not only by directly changing the $\dot{O}$H and HO 2 radical pools via fuel-specific reactions, but also through intermolecular interactions where participating intermediates can be produced and consumed by various sub-chemistries. Disabling the intermolecular interactions in carbonated species between ethanol and FGF-LLNL sub-chemistries leads to somewhat slower ITHR evolution and lower ignition reactivity. The role of the individual intermolecular interaction is also characterized using the proposed approach. Lastly, implications of intermolecular interactions for future studies that aim to improve model performance are highlighted, where it is found that further investigations on the core C0-C4 chemistry are needed for developing highly accurate chemistry models for complex fuel blends.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

Machine learning and deep learning for mineralogy interpretation and CO 2 saturation estimation in geological carbon Storage: A case study in the Illinois Basin

Carbon capture and storage (CCS) is a promising approach to simultaneously maintaining energy security and reducing carbon dioxide (CO 2 ) emissions under the current energy portfolio that is dominated by fossil fuel energy. Pre-injection formation characterization and post-injection CO 2 monitoring are two critical tasks to guarantee storage efficiency in CCS. The CCS projects in the Illinois Basin, the first large-scale CO 2 injection into saline aquifers in the United States, employed conventional and the latest pulsed neutron logging (PNL) tools for mineralogy interpretation and CO 2 saturation estimation, which provide valuable references for future CCS projects. Because of the inherent fuzziness of petrophysical measurements and complex subsurface heterogeneity, interpreting well-logging data is time-consuming, and its accuracy can be user-biased. In recent years, data-driven methods have been widely used to capture the non-linear patterns between input features and interpretation results. This work applied and evaluated four commonly used machine learning (ML) models, including ridge regression (RR), random forest (RF), gradient boosting regression (GBR), support vector regression (SVR), and one deep learning (DL) model, the artificial neural network (ANN). We optimized the hyperparameters of the four ML models and the DL model using the simulated annealing algorithm and the grid search strategy, respectively. The input features of the mineralogy interpretation models were eleven conventional well-logging parameters, and the label data (i.e., ground truth) were the porosity and volumetric fractions of six minerals, including quartz, feldspar, dolomite, calcite, clay, and iron minerals. The results demonstrated that the GBR and RF models were superior in predicting volumetric fractions of minerals and porosity; label data with low coefficient of variation (CV) values tended to yield better performance. For CO 2 saturation estimation, the RF was the best-performing model, followed by SVR, ANN, GBR, and RR. Furthermore, we conducted feature importance ranking using the permutation importance algorithm and found that the formation sigma and well pressure were the most important features in this study. In conclusion, the study of CCS projects in the Illinois Basin bridges the gap between the limited knowledge and understanding of geological carbon storage and the increasing demand for reliable, cost-effective, and sustainable energy solutions.

58 GEOSCIENCES↗
DOE OSTI2022

Waveform processing using neural network algorithms on the front-end electronics

In a multi-channel radiation detector readout system, waveform sampling, digitization, and raw data transmission to the data acquisition system constitute a conventional processing chain. The deposited energy on the sensor is estimated by extracting peak amplitudes, area under pulse envelopes from the raw data, and starting times of signals or time of arrivals. However, such quantities can be estimated using machine learning algorithms on the front-end Application-Specific Integrated Circuits (ASICs), often termed as “edge computing”. Edge computation offers enormous benefits, especially when the analytical forms are not fully known or the registered waveform suffers from noise and imperfections of practical implementations. In this work, we aim to predict peak amplitude from a single waveform snippet whose rising and falling edges containing only 3 to 4 samples. We thoroughly studied two well-accepted neural network algorithms, Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN) by varying their model sizes. Further, to better fit front-end electronics, neural network model reduction techniques, such as network pruning methods and variable-bit quantization approaches, were also studied. By combining pruning and quantization, our best performing model has the size of 1.5 KB, reduced from 16.6 KB of its full model counterpart. It can reach mean absolute error of 0.034 comparing to that of a naive baseline of 0.135. Such parameter-efficient and predictive neural network models established feasibility and practicality of their deployment on front-end ASICs.

47 OTHER INSTRUMENTATION↗
DOE OSTI2025

Data imbalance in drug response prediction: multi-objective optimization approach in deep learning setting

Abstract Drug response prediction (DRP) methods tackle the complex task of associating the effectiveness of small molecules with the specific genetic makeup of the patient. Anti-cancer DRP is a particularly challenging task requiring costly experiments as underlying pathogenic mechanisms are broad and associated with multiple genomic pathways. The scientific community has exerted significant efforts to generate public drug screening datasets, giving a path to various machine learning models that attempt to reason over complex data space of small compounds and biological characteristics of tumors. However, the data depth is still lacking compared to application domains like computer vision or natural language processing domains, limiting current learning capabilities. To combat this issue and improves the generalizability of the DRP models, we are exploring strategies that explicitly address the imbalance in the DRP datasets. We reframe the problem as a multi-objective optimization across multiple drugs to maximize deep learning model performance. We implement this approach by constructing Multi-Objective Optimization Regularized by Loss Entropy loss function and plugging it into a Deep Learning model. We demonstrate the utility of proposed drug discovery methods and make suggestions for further potential application of the work to achieve desirable outcomes in the healthcare field.

Biochemistry & Molecular Biology↗
DOE OSTI2020

NEON AOP foliar trait maps, maps of model uncertainty estimates, and conifer map, East River, CO 2018

This data package contains mapped trait estimates and their uncertainties, and conifer map, for the National Ecological Observatory Network's Airborne Observation Platform survey data acquired over the Upper East River, Colorado in 2018. For full details, please see associated reference. in brief, trait models were developed independently for needle and non-needle leaf species using partial least squares regression (PLSR) using ground data from additional datasets: doi:10.15485/1618130, doi:10.15485/1618132, and doi:10.15485/1631278, merged with extracted spectral data from doi:10.15485/1618131. We separated vegetated pixels into needle and non-needle classes in order to generate a classification map based on the spectral differences between these leaf types (conifer.tif). We trained a deep learning model with custom architecture, detailed in Chadwick et al. In Press. The model performed with 0.998 true positive rate and 0.982 true negative rate, with ‘positives’ being non-needle identification. We then utilized PLSR to generate models of foliar traits for each leaf type. So that we could also map uncertainty in these predictions, we generated ten different models for needle and non-needle leaf species using different testing holdout sets of discrete sites. Each of these models was developed with a 100-fold cross validation procedure that utilized a 70% training set and 30% validation set with each fold, and then assessed based on the 10% of testing sites that were not included in that model’s development. The mean predicted value across the 10 models is used for the trait estimate in each pixel across the study area. The models are applied according to the leaf type designation in the conifer.tif map. The errors are the standard deviation across the 10 different models developed, with high error suggesting instability in model prediction and areas where values may not be reliable for ecological inference. These maps are only applied to areas with a NDVI > 0.5 to exclude non-vegetated areas. Shade masks could be applied to these data (doi:10.15485/1618131), but have not been for this data package. These data are also available on Google Earth Engine: https://code.earthengine.google.com/?asset=users/kdc/ER_NEON

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Semi-supervised permutation invariant particle-level anomaly detection

The development of analysis methods to distinguish potential beyond the Standard Model phenomena in a model-agnostic way can significantly enhance the discovery reach in collider experiments. However, the typical machine learning (ML) algorithms employed for this task require fixed length and ordered inputs that break the natural permutation invariance in collision events. To address this, a semi-supervised anomaly detection tool is presented that takes a variable number of particle-level inputs and leverages a signal model to encode this information into a permutation invariant, event-level representation via supervised training with a Particle Flow Network (PFN). Data events are then encoded into this representation and given as input to an autoencoder for unsupervised ANomaly deTEction on particLe flOw latent sPacE (ANTELOPE), classifying anomalous events based on a low-level and permutation invariant input modeling. Performance of the ANTELOPE architecture is evaluated on simulated samples of hadronic processes in a high energy collider experiment, showing good capability to distinguish disparate models of new physics.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

Multi-module-based CVAE to predict HVCM faults in the SNS accelerator

We present a multi-module framework based on Conditional Variational Autoencoder (CVAE) to detect anomalies in the power signals coming from multiple High Voltage Converter Modulators (HVCMs). We condition the model with the specific modulator type to capture different representations of the $\mathcal{normal}$ waveforms and to improve the sensitivity of the model to identify a specific type of fault when we have limited samples for a given module type. We studied several Artificial Neural Network (ANN) architectures for our CVAE model and evaluated the model performance by looking at their loss landscape for stability and generalization. Our results for the Spallation Neutron Source (SNS) experimental data show that the trained model generalizes well to detecting multiple fault types for several HVCM module types. The results of this study can be used to improve the HVCM reliability and overall SNS uptime.

43 PARTICLE ACCELERATORS↗
DOE OSTI2024

Merged Observatory Data Files (MODFs): an integrated observational data product supporting process-oriented investigations and diagnostics

A large and ever-growing body of geophysical information is measured in campaigns and at specialized observatories as a part of scientific expeditions and experiments. These collections of observed data include many essential climate variables (as defined by the Global Climate Observing System) but are often distinguished by a wide range of additional non-routine measurements that are designed to not only document the state of the environment but also the drivers that contribute to that state. These field data are used not only to further understand environmental processes through observation-based studies but also to provide baseline data to test model performance and to codify understanding to improve predictive capabilities. To address the considerable barriers and difficulty in utilizing these diverse and complex data for observation–model research, the Merged Observatory Data File (MODF) concept has been developed. A MODF combines measurements from multiple instruments into a single file that complies with well-established data format and metadata practices and has been designed to parallel the development of corresponding Merged Model Data Files (MMDFs). Using the MODF and MMDF protocols will facilitate the evolution of model intercomparison projects into model intercomparison and improvement projects by putting observation and model data “on the same page” in a timely manner. The MODF concept was developed especially for weather forecast model studies in the Arctic. The surprisingly complex process of implementing MODFs in that context refined the concept itself. Thus, this article explains the concept of MODFs by providing details on the issues that were revealed and resolved during that first specific implementation. Detailed instructions are provided on how to make MODFs, and this article can be considered a MODF creation manual.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Assessing time-dependent temperature profile predictions using reduced transport models for high performing NSTX plasmas

Time-dependent, predictive simulations were performed with the 1.5D tokamak integrated modeling code TRANSP on a large set of well-analyzed, high performing discharges from the National Spherical Torus Experiment (NSTX) in order to evaluate how well modern reduced transport models can reproduce experimentally observed temperature profiles in spherical tokamaks. Overall, it is found that simulations using the Multi-Mode Model (MMM) more consistently agree with the NSTX observations than those using the Trapped Gyro-Landau Fluid (TGLF) model, despite TGLF requiring orders of magnitude greater computational cost. When considering all examined discharges, MMM has median overpredictions of electron temperature (T e ) and ion temperature (T i ) profiles of 28% and 27%, respectively, relative to the experiment. TGLF overpredicts T e by 46%, with much larger variance than MMM, and underpredicts T i by 25%. As the ratio of kinetic to magnetic field pressure (β) is increased across NSTX discharges, TGLF predicts lower T e and significant flattening of the T i profile, conflicting with NSTX observations. When using an electrostatic version of TGLF, both T e and T i are substantially overpredicted, underscoring the importance of electromagnetic turbulence in the high β spherical tokamak regime. Additionally, calculations with neural net surrogate models for TGLF were performed outside of TRANSP with a time slice flux matching transport solver, finding better agreement with experiment than the TRANSP simulations, highlighting the impact of different transport solvers and simulation techniques. Furthermore, the reasonable agreement with experiment of temperature profiles predicted by MMM motivates a more detailed examination of the sensitivities of the TRANSP simulations with MMM to different NSTX plasma regimes in a companion paper (Lestz et al Plasma Phys. Control. Fusion 67 105030), in preparation for self-consistent, time-dependent predictive modeling of NSTX-U scenarios.

Lestz, J. B. [General Atomics, San Diego, CA (Unit↗
DOE OSTI2023

Combining multitask and transfer learning with deep Gaussian processes for autotuning-based performance engineering

We combine deep Gaussian processes (DGPs) with multitask and transfer learning for the performance modeling and optimization of HPC applications. Deep Gaussian processes merge the uncertainty quantification advantage of Gaussian processes (GPs) with the predictive power of deep learning. Multitask and transfer learning allow for improved learning efficiency when several similar tasks are to be learned simultaneously and when previous learned models are sought to help in the learning of new tasks, respectively. A comparison with state-of-the-art autotuners shows the advantage of our approach on two application problems. In this article, we combine DGPs with multitask and transfer learning to allow for both an improved tuning of an application parameters on problems of interest but also the prediction of parameters on any potential problem the application might encounter.

97 MATHEMATICS AND COMPUTING↗