Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Transfer Learning Model”

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 433 records · Page 24
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Physics-Informed Neural Network (PINN) Prediction of Mixed Mass-Heat-Crystallization Limited Methane Hydrate Formation and Dissociation in Micro-Confinement

The creation and use of Physics-Informed Neural Networks (PINNs) for simulating the dynamics of methane hydrate formation and dissociation will be presented. The PINN framework's main benefit is its capacity to impose physical consistency with only a partial comprehension of the governing equations. This makes the algorithm especially useful for systems with little experimental evidence or a lack of theoretical knowledge. A strong basis for forecasting methane hydrate behavior over the verified operating ranges of 30.0-80.9 bar pressure and 1.0-4.0 K sub-cooling conditions is provided by the combination of conductive heat transfer equations and mixed mass-transfer–crystallization kinetics. PINNs were more accurate at predicting the mixed mass-heat-crystallization limited kinetics than conventional Artificial Neural Networks (ANNs), demonstrating remarkable predictive accuracy for methane hydrate production over the ANN model. The efficiency of incorporating physical limitations from first principles into machine learning frameworks for methane hydrate crystallizations is reinforced by these findings. For hydrate-related applications in energy generation, carbon sequestration, and climate modelling, our study establishes PINNs as a computational tool that is both scalable and efficient. The proven capacity to close the gap between conventional physics-based simulations and solely data-driven models creates new opportunities for expedited hydrate research and practical applications.

Hartman, Ryan L [NYU Tandon School of Engineering]↗
DOE OSTI2024

High-Fidelity and High-Performance Computational Simulations for Rapid Design Optimization of Sulfur Thermal Energy Storage

Industrial process heating (IPH) accounts for approximately 70% of US manufacturing energy use and is primarily produced by fossil fuel combustion. Approximately 1500 TWht (approximately 60%) of IPH demand is in the temperature range of 100-300. Industrial applications in this temperature range include drying, hydrothermal processing, thermal enhanced oil recovery, food and beverage, bioethanol production, etc. Cost-effective thermal energy storage (TES) that increases the utilization of waste and renewable heat (solar, geothermal, etc.) could provide significant energy savings and reliable heat sources, decrease emissions, and increase US manufacturing competitiveness through reductions in fuel consumption. TES development has historically been dominated by technologies suitable for deployment with concentrating solar power (CSP). State-of-the-art thermal storage deployed commercially with power tower CSP plants uses a 60%/40% NaNO3/KNO3 molten salt and operates between temperatures of approximately 280 degrees Celsius and 570 degrees Celsius using a two-tank configuration. However, these nitrate salts are unsuitable for operation outside of this temperature range due to a high freezing point of approximately 220 degrees Celsius, and limits on high-temperature salt stability and corrosion resistance of containment alloys. Other materials being investigated for TES include those based on: (1) sensible energy storage (various molten salt compositions, inert solid particles, rocks or pebble beds, sulfur, water, concrete, graphite, etc.), (2) latent energy storage in materials that undergo solid-liquid phase change at relevant temperatures (organic materials for low-temperature applications, inorganic salts and/or metals for high-temperature applications), or (3) thermochemical energy storage (hydrides, hydroxides, carbonates, metal oxides, etc.). The application temperature and challenges pertaining to storage material and/or containment cost, energy density, long-term thermal and cyclic stability, and charge/discharge heat transfer effectiveness drive material selection for a given IPH or electricity generation application. Sulfur is a cheap commodity at $80/ton compared to $1100 - 1300/ton for conventional salts. When using a metric of storage cost per kWh, sulfur costs around 2-3 $/kWh. Previous sulfur TES development focused on high temperature (>600 degrees) concentrated solar power applications with sulfur encapsulated in pipes and flow of gaseous HTF (air) in the shell side. However, for lower-temperature IPH applications in the range of approximately 100-300 degrees Celsius Element 16 adopted a compact and scalable TES design with molten sulfur in the shell and HTF pipes submerged in the molten sulfur bath. The low-cost molten sulfur TES for dispatchable IPH has deployment potential for broad applications. The spatial and temporal evolution of the HTF and sulfur temperature is critical to the TES system performance, and thus detailed modeling can improve understanding of the performance and facilitate design improvements. Using high performance computing and computational fluid dynamics (CFD) a low-cost molten sulfur thermal energy storage (TES) system for industrial process heating (IPH) applications was developed. The unique challenges in CFD modeling of sulfur TES are the sharp property changes of sulfur relevant to the working temperatures. Above 159, liquid sulfur undergoes polymerization, and the viscosity of sulfur rapidly increases by several orders of magnitude between 159 degrees Celsius and 188 degrees Celsius, followed by a decrease in viscosity beyond 188 degrees Celsius due to thermal bound dissociation. In addition, various concentrations of H2S impurities can also modify sulfur viscosity. This numerical challenge is especially relevant to transient simulation of the sulfur TES charging and discharging processes as the extreme property variations limit the applicability of traditional heat transfer correlations. Transient CFD simulations including the temperature-dependent sulfur properties and geometric complexity of the TES design were used to predict the effect of natural convection during charging and discharging on the heat transfer process, sulfur temperature uniformity, charge/discharge rates, and performance of the storage devices. The CFD model was validated with experimental results for a full charge and discharge cycle. The work will show 3D and 2D simulation comparisons aimed to facilitate rapid design iterations and a machine learning based design optimization approach.

CFD↗
DOE OSTI2023

BuildingsBench: A Benchmark for Universal Building Load Forecasting [SWR-23-51]

The residential and commercial building stock in the United States is responsible for a significant percentage of energy consumption and greenhouse gas emissions. Electrification of end-uses, as well as decarbonizing the electrical grid through renewable energy sources such as solar and wind, constitutes the pathway to zero-emission buildings. Forecasting day-ahead building energy consumption is an integral part of this solution. Currently, specialized forecasting models are hand-made for each individual building, which is time-consuming, expensive, and leads to duplicated efforts. BuildingsBench is a Python software framework for training and comparing generalized machine learning models for universal building load forecasting. This challenge tasks a single foundational model to generalize its forecasts for a wide variety of buildings, across geographic regions, building types, weather patterns, and more. This software provide code for pre-training such models and subsequently evaluating their performance on a suite of hundreds of diverse real and synthetic buildings. BuildingsBench is a platform for: - Large-scale pretraining with the synthetic Buildings-900K dataset for short-term load forecasting (STLF). Buildings-900K is statistically representative of the entire U.S. building stock and is extracted from the NREL End-Use Load Profiles database. - Benchmarking on two tasks evaluating generalization: zero-shot STLF and transfer learning for STLF. We provide an index-based PyTorch Dataset for large-scale pretraining, easy data loading for multiple real building energy consumption datasets as PyTorch Tensors or Pandas DataFrames, simple (persistence) to advanced (transformer) baselines, metrics management, and more.

Emami, Patrick↗
DOE OSTI2025

You Only Look Once v5 and Multi-Template Matching for Small-Crack Defect Detection on Metal Surfaces

This paper compares the performance of Deep Learning (DL) and multi-template matching (MTM) models for detecting small defects. DL models extract distinguishing features of objects but require a large dataset of images. In contrast, alternative computer vision techniques like MTM need a relatively small dataset. The lack of large datasets for small metal-surface defects has inhibited the adoption of automation in small-defect detection in remanufacturing settings. This motivated this preliminary study to compare template-based approaches, like MTM, with feature-based approaches, such as DL models, for small-defect detection on an initial laboratory and remanufacturing industry dataset. This study used You Only Look Once v5 (YOLOv5) as the DL model and compared its performance against the MTM model for small-crack detection. The findings of our preliminary investigation are as follows: (i) YOLOv5 demonstrated higher performance than MTM in detecting small cracks; (ii) an extra-large variant of YOLOv5 outperformed a small-size variant; (iii) the size and object variety of the data are crucial in achieving robust pre-trained weights for use in transfer learning; and (iv) enhanced image resolution contributes to precise object detection.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2020

Improving radiograph analysis throughput through transfer learning and object detection

SIGN Fracture Care International partners with surgeons in low-resource hospitals worldwide to provide access to effective orthopedic care by donating educational materials and innovatively designed surgical implants. Over two decades, SIGN’s Online Surgical Database (SOSD) has grown to contain over 500,000 medical images, with radiographs holding the majority share. One challenge in working with hospitals worldwide is that both the radiographs uploaded to the SOSD and the data entry accompanying the uploads vary in quality. To improve the accuracy of data in the SOSD, we trained a model to detect surgical implants in radiographs. We first developed a tool to automatically detect radiographs, then trained an object detection model to determine the number and placement of surgical implants visible in the radiograph. Active learning was used to generate a training set containing 2,510 radiographs with screws, nails, and plates labeled by bounding boxes. Training a model to simultaneously recognize all three classes of implants gave a low average precision (AP) for the plate class, likely due to the low number of plate instances in our training set and the large variety of surgical plates used by SIGN-partnered surgeons. Applying standard image augmentation techniques to increase the plate count in our training set did not appreciably increase the AP of plate detection. To improve plate detection, we redrew the bounding boxes to account for correlations between the screw and plate classes. Training one model to detect nails and screws and a separate model to detect plates increased the AP of plate detection by 78.8 percentage points. The AP of each class was 80.7% for screws, 93.6% for nails, and 92.6% for plates; meanwhile, the sensitivity was 92% for screws, 86% for nails, and 81% for plates. We show that object detection methods can be used to detect surgical implants in radiographs of varying quality; however, the detection ability is dependent on the type of implant, and some implants, in our case plates, must be treated differently than others. Such tools can improve the throughput of radiograph analysis, assisting physicians and surgeons with the treatment of bone fractures.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

Estimating cluster masses from SDSS multiband images with transfer learning

ABSTRACT The total masses of galaxy clusters characterize many aspects of astrophysics and the underlying cosmology. It is crucial to obtain reliable and accurate mass estimates for numerous galaxy clusters over a wide range of redshifts and mass scales. We present a transfer-learning approach to estimate cluster masses using the ugriz-band images in the SDSS Data Release 12. The target masses are derived from X-ray or SZ measurements that are only available for a small subset of the clusters. We designed a semisupervised deep learning model consisting of two convolutional neural networks. In the first network, a feature extractor is trained to classify the SDSS photometric bands. The second network takes the previously trained features as inputs to estimate their total masses. The training and testing processes in this work depend purely on real observational data. Our algorithm reaches a mean absolute error (MAE) of 0.232 dex on average and 0.214 dex for the best fold. The performance is comparable to that given by redMaPPer, 0.192 dex. We have further applied a joint integrated gradient and class activation mapping method to interpret such a two-step neural network. The performance of our algorithm is likely to improve as the size of training data set increases. This proof-of-concept experiment demonstrates the potential of deep learning in maximizing the scientific return of the current and future large cluster surveys.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2023

CEGANN: CRYSTAL EDGE GRAPH ATTENTION NEURAL NETWORK

SF-22-156 Machine learning (ML) models and applications in materials design and discovery typically involve the use of feature representations or descriptors followed by a learning algorithm that maps them to user desired properties of interest. Most popular mathematical formulation-based descriptors are not unique across atomic environments and suffer from transferability issues across different application domains and/or material classes. The CEGANN code provides a unified interface to facilitate material characterization across materials across multiple scales (from atomic to mesoscale) and diverse classes of materials ranging from metals oxides, non-metals, and even hierarchical materials such as zeolites and semi ordered materials such as mesophases. CEGANN implements a Graph Attention Network (GAT) type convolution architecture. The details of network architecture can be found in the paper https://doi.org/10.48550/arXiv.2207.10168. The software comes with pretrained examples and dataset for the classification of the following representative systems: (1) Structure-level representation such as space group (2) Structural dimensionality (e.g., bulk, 2D, clusters etc.) (3) Grain boundary identification (4) Nucleation and growth of a zeolite polymorph (5) Characterization of binary mesophases and their phase transitions (6) Growth of ice. The code is written in python programming language.

CHAN, HENRYT↗
DOE OSTI2024

Concept Study of Robotic Camera-Based Foreign Object Detection for EV Wireless Charging

Wireless charging of an electric vehicle (EV) is an emerging charging technology promising convenient, autonomous, and highly efficient EV charging without requiring heavy gauge cables. However, due to the strong electromagnetic field created by this process that surrounds the wireless charger, the presence of foreign objects can detrimentally interact with it, thus affecting wireless power transfer (WPT) performance or leading to harmful and unwanted safety risks. This paper presents the results for a concept study on a robotic camera-based foreign object detection (FOD) system, as a supplement to the industry-existing overlapped FOD coil array method, for EV wireless charging. A Raspberry PI 4 control board and compatible Raspberry PI Camera Module 2 are used to implement camera-based object detection. The FOD program was developed using a state-of-the-art deep learning object detection model with the OpenCV and Pytorch library and is compatible with camera module hardware. A dry-run test with Raspberry PI and a camera module was conducted and the preliminary FOD function was verified. The feasibility assessment is also validated by comparing the performance of five existing state-of-the-art deep learning object detection models for vehicles, animals, persons, and metals subsets, respectively. Satisfactory performance on the benchmark datasets is observed by the tests, but further improvements are needed in future work when detecting small-sized metallic objects. A programable robotic car is also under development as ongoing work for carrying the Raspberry PI and camera module while moving for the maintenance process.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2020

Automatic Classification of Biological Targets in a Tidal Channel Using a Multibeam Sonar

Multibeam sonars are widely used for environmental monitoring of fauna at marine renewable energy sites. However, they can rapidly accrue vast volumes of data, which poses a challenge for data processing. Here, using data from a deployment in a tidal channel with peak currents of 1–2 m s –1 , we demonstrate the data-reduction benefits of real-time automatic classification of targets detected and tracked in multibeam sonar data. First, we evaluate classification capabilities for three machine learning algorithms: random forests, support vector machines, and k-nearest neighbors. For each algorithm, a hill-climbing search optimizes a set of hand-engineered attributes that describe tracked targets. Here, the random forest algorithm is found to be most effective—in postprocessing, discriminating between biological and nonbiological targets with a recall rate of 0.97 and a precision of 0.60. In addition, 89% of biological targets are correctly classified as either seals, diving birds, fish schools, or small targets. Model dependence on the volume of training data is evaluated. Second, a real-time implementation of the model is shown to distinguish between biological targets and nonbiological targets with nearly the same performance as in postprocessing. From this, we make general recommendations for implementing real-time classification of biological targets in multibeam sonar data and the transferability of trained models.

16 TIDAL AND WAVE POWER↗
DOE OSTI2021

Watch and learn—a generalized approach for transferrable learning in deep neural networks via physical principles

Transfer learning refers to the use of knowledge gained while solving a machine learning task and applying it to the solution of a closely related problem. Such an approach has enabled scientific breakthroughs in computer vision and natural language processing where the weights learned in state-of-the-art models can be used to initialize models for other tasks which dramatically improve their performance and save computational time. Here we demonstrate an unsupervised learning approach augmented with basic physical principles that achieves fully transferrable learning for problems in statistical physics across different physical regimes. By coupling a sequence model based on a recurrent neural network to an extensive deep neural network, we are able to learn the equilibrium probability distributions and inter-particle interaction models of classical statistical mechanical systems. Our approach, distribution-consistent learning, DCL, is a general strategy that works for a variety of canonical statistical mechanical models (Ising and Potts) as well as disordered interaction potentials. Using data collected from a single set of observation conditions, DCL successfully extrapolates across all temperatures, thermodynamic phases, and can be applied to different length-scales. This constitutes a fully transferrable physics-based learning in a generalizable approach.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Distilling particle knowledge for fast reconstruction at high-energy physics experiments

Knowledge distillation is a form of model compression that allows artificial neural networks of different sizes to learn from one another. Its main application is the compactification of large deep neural networks to free up computational resources, in particular on edge devices. In this article, we consider proton-proton collisions at the High-Luminosity Large Hadron Collider (HL-LHC) and demonstrate a successful knowledge transfer from an event-level graph neural network (GNN) to a particle-level small deep neural network (DNN). Our algorithm, DistillNet, is a DNN that is trained to learn about the provenance of particles, as provided by the soft labels that are the GNN outputs, to predict whether or not a particle originates from the primary interaction vertex. The results indicate that for this problem, which is one of the main challenges at the HL-LHC, there is minimal loss during the transfer of knowledge to the small student network, while improving significantly the computational resource needs compared to the teacher. This is demonstrated for the distilled student network on a CPU, as well as for a quantized and pruned student network deployed on a field programmable gate array. Our study proves that knowledge transfer between networks of different complexity can be used for fast artificial intelligence (AI) in high-energy physics that improves the expressiveness of observables over non-AI-based reconstruction algorithms. Such an approach can become essential at the HL-LHC experiments, e.g. to comply with the resource budget of their trigger stages.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2020

Robust Online Sequential RVFLNs for Data Modeling of Dynamic Time-Varying Systems with Application of an Ironmaking Blast Furnace

In a world where the increasing complexity of modern industrial processes brings difficulties for accurate mathematical modeling, taking advantage of data has become an efficient solution to complex dynamic process modeling issue. In this paper, we develop a novel robust online sequential version of random vector functional-link networks (RVFLNs) for data-driven modeling of dynamic time-varying system and applied it in a blast furnace (BF) ironmaking process. First, to overcome the time-varying dynamics of process and to enable the RVFLNs to learn online with avoiding data saturation, an improved online sequential version of RVFLNs (OS-RFVLNs) is first presented by online sequential learning with forgetting factor. This improved OS-RVFLNs algorithm is not only suitable for the real-time and large data transfer situation, but also can adjust the sensitivity of the algorithm to different samples with the help of the introduced forgetting factor. Second, since the output weights of the improved OS-RVFLNs as well as other RVFLNs algorithms are obtained by the least squares approach, a robustness problem may occur when the training dataset is contaminated with various outliers. To solve this problem, a Cauchy distribution weighted M-estimator is introduced to improve the robustness of the improved OS- RVFLNs. For this proposed robust OS-RVFLNs (R-OS- RVFLNs), since the weights of different outlier data are properly determined by the Cauchy distribution function, their corresponding contribution on modeling can be properly distinguished. Thus robust and better modeling results can be achieved. Experiments using actual industrial data of BF ironmaking process and comparative studies have demonstrated that the proposed method produces a better estimation accuracy and stronger robustness than other methods.

Blast furnace (BF), Modelling, Dynamic systems↗
DOE OSTI2023

Data-driven prediction of α IIb β 3 integrin activation paths using manifold learning and deep generative modeling

The integrin heterodimer is a transmembrane protein critical for driving cellular process and is a therapeutic target in the treatment of multiple diseases linked to its malfunction. Activation of integrin involves conformational transitions between bent and extended states. Some of the conformations that are intermediate between bent and extended states of the heterodimer have been experimentally characterized, but the full activation pathways remain unresolved both experimentally due to their transient nature and computationally due to the challenges in simulating rare barrier crossing events in these large molecular systems. An understanding of the activation pathways can provide new fundamental understanding of the biophysical processes associated with the dynamic interconversions between bent and extended states and unveil new putative therapeutic targets. In this work, we apply nonlinear manifold learning to coarse-grained molecular dynamics simulations of bent, extended, and two intermediate states of αI I b β3 integrin to learn a low-dimensional embedding of the configurational phase space. We then train deep generative models to learn an inverse mapping between the low-dimensional embedding and high-dimensional molecular space and use these models to interpolate the molecular configurations constituting the activation pathways between the experimentally characterized states. Furthermore, this work furnishes plausible predictions of integrin activation pathways and reports a generic and transferable multi-scale technique to predict transition pathways for biomolecular systems.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Deep reinforcement learning control of hydraulic fracturing

Hydraulic fracturing is a technique to extract oil and gas from shale formations, and obtaining a uniform proppant concentration along the fracture is key to its productivity. Recently, various model predictive control schemes have been proposed to achieve this objective. But such controllers require an accurate and computationally efficient model which is difficult to obtain given the complexity of the process and uncertainties in the rock formation properties. In this article, we design a model-free data-based reinforcement learning controller which learns an optimal control policy through interactions with the process. Deep reinforcement learning (DRL) controller is based on the Deep Deterministic Policy Gradient algorithm that combines Deep-Q-network with actor-critic framework. In addition, we utilize dimensionality reduction and transfer learning to quicken the learning process. We show that the controller learns an optimal policy to obtain uniform proppant concentration despite the complex nature of the process while satisfying various input constraints.

42 ENGINEERING↗
DOE OSTI2022

Machine learning-based optimization of air-cooled heat sinks

Machine learning-based models using Artificial Neural Network (ANN) and greedy search algorithm are used to optimize air-cooled parallel plate-finned heat sinks (PPFHSs) subjected to laminar flow over an extensive range of design parameters. Here, the thermal and hydraulic performances of PPFHSs are represented by heat transfer coefficient (h) and pressure drop (ΔP), respectively. Optimization objectives for PPFHS designs can vary from industry to industry depending on their design priorities. The present study proposes a novel and generalized optimization method that defines practical optimization objectives and provides an accurate optimization process to design effective PPFHSs for a wide range of industrial applications with different design requirements. Three optimization objectives are presented in this study: (i) the largest h PΔ, (ii) the largest h within a specified maximum allowed flow rate, and (iii) the lowest weight that maximizes h for operation within the maximum allowed flow rate. While the shortcoming of the first objective is demonstrated, the other two objectives are found to be suitable for designing effective heat sinks (HSs) across different applications. Results suggest a promising trend from the third objective to develop HSs with ~ 37-68% lower weight, 80-85% reduced ΔP, and negligible penalty in h compared with optimized HSs obtained from the second objective. However, since the third objective leads to HSs with thinner fins, structural analysis should be performed to ensure reliable operation of the HSs.

42 ENGINEERING↗
DOE OSTI2023

Deconvoluting experimental decay energy spectra: The O 26 case

In nuclear reaction experiments, the measured decay energy spectra can give insights into the shell structure of decaying systems. However, extracting the underlying physics from the measurements is challenging due to detector resolution and acceptance effects. The Richardson-Lucy (RL) algorithm, a deblurring method that is commonly used in optics and has proven to be a successful technique for restoring images, was applied to our experimental nuclear physics data. The only inputs to the method are the observed energy spectrum and the detector's response matrix also known as the transfer matrix. We demonstrate that the technique can help access information about the shell structure of particle-unbound systems from the measured decay energy spectrum that is not immediately accessible via traditional approaches such as χ-square fitting. For a similar purpose, we developed a machine learning model that uses a deep neural network (DNN) classifier to identify resonance states from the measured decay energy spectrum. We tested the performance of both methods on simulated data and experimental measurements. Then, we applied both algorithms to the decay energy spectrum of 26 O → 24 O + n + n measured via invariant mass spectroscopy. Here, the resonance states restored using the RL algorithm to deblur the measured decay energy spectrum agree with those found by the DNN classifier. Both deblurring and DNN approaches suggest that the raw decay energy spectrum of 26 O exhibits three peaks at approximately 0.15 MeV, 1.50 MeV, and 5.00 MeV, with half-widths of 0.29 MeV, 0.80 MeV, and 1.85 MeV, respectively.

Spectrometers & spectroscopic techniques↗
DOE OSTI2024

Efficient Parallelization of Irregular Applications on GPU Architectures

With the enlarging computation capacity of general Graphics Processing Units (GPUs), leveraging GPUs to accelerate parallel applications has become a critical topic in academia and industry. However, a wide range of irregular applications with the computation-/memory-intensive nature cannot easily achieve high GPU utilization. The challenges mainly involve the following aspects: first, data dependence leads to coarse-grained kernel and inefficient parallelism; second, heavy GPU memory usage may cause frequent memory evictions and extra overhead of I/O; third, specific computation patterns produce memory redundancies; last, workload balance and data reusability conjunctly benefit the overall performance, but there may exist a dynamic trade-off between them. Targeting these challenges, this dissertation proposes multiple optimizations to accelerate two real-world applications: many-body correlation functions to simulate nuclear physics in a large-scale scientific system; the other is the eALS-based matrix factorization recommendation system. To accelerate the calculations of many-body correlation functions, this dissertation presents three frameworks in GPU memory management and multi-GPU scheduling. Firstly, an optimized systematic GPU memory management framework, MemHC, utilizes a series of new memory reduction designs in GPU memory allocation, CPU/GPU communications, and GPU memory oversubscription. Secondly, an enhanced multi-GPU scheduling framework, MICCO, particularly by taking both data dimension (e.g., data reuse and data eviction) and computation dimension into account. MICCO designs a heuristic scheduling algorithm and a machine learning-based regression model to generate the optimal settings of a proposed new concept to manage the trade-off. Thirdly, a locality-aware multi-GPU scheduling framework. This scheduler leverages pipeline batch generation with a looking-ahead strategy by building local dependency graphs for memory transfer reduction and better data reuse, achieving up to 79.92% memory cost reduction and 1.67x speedup. To parallelize the eALS-based recommendation system, this dissertation proposes an efficient CPU/GPU heterogeneous recommendation system, HEALS. HEALS employs newly designed architecture-adaptive data formats to achieve load balance and good data locality on CPU and GPU. To mitigate the data dependence, HEALS presents a CPU/GPU collaboration model for both task parallelism and data parallelism with multiple kernel computation optimizations. In summary, this dissertation efficiently accelerates two typical irregular applications on GPUs by building four frameworks, including CPU/GPU collaboration, GPU memory management, and multi-GPU scheduling.

Wang, Qihan↗
DOE OSTI2021

Machine-learning-based dynamic-importance sampling for adaptive multiscale simulations

Multiscale simulations are a well-accepted way to bridge the length and time scales required for scientific studies with the solution accuracy achievable through available computational resources. Traditional approaches either solve a coarse model with selective refinement or coerce a detailed model into faster sampling, both of which have limitations. Here, we present a paradigm of adaptive, multiscale simulations that couple different scales using a dynamic-importance sampling approach. Our method uses machine learning to dynamically and exhaustively sample the phase space explored by a macro model using microscale simulations and enables an automatic feedback from the micro to the macro scale, leading to a self-healing multiscale simulation. As a result, our approach delivers macro length and time scales, but with the effective precision of the micro scale. Our approach is arbitrarily scalable as well as transferable to many different types of simulations. Overall, our method made possible a multiscale scientific campaign of unprecedented scale to understand the interactions of RAS proteins with a plasma membrane in the context of cancer research running over several days on Sierra, which is currently the second-most-powerful supercomputer in the world.

59 BASIC BIOLOGICAL SCIENCES↗