Search NASA⌕ Search

SEARCH · Search NASA

Results for “parallel machine learning algorithms”

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.

143 records · Page 8

Stabilization of the 81-channel coherent beam combination using machine learning

We develop a rapidly converging algorithm for stabilizing a large channel-count diffractive optical coherent beam combination. An 81-beam combiner is controlled by a novel, machine-learning based, iterative method to correct the optical phases, operating on an experimentally calibrated numerical model. A neural-network is trained to detect phase errors based on interference pattern recognition of uncombined beams adjacent to the combined one. Due to the non-uniqueness of solutions in the full space of possible phases, the network is trained within a limited phase perturbation/error range. This also reduces the number of samples needed for training. Simulations have proven that the network can converge in one step for small phase perturbations. When the trained neural-network is applied to a realistic case of 360 degree full range, an iterative scheme exploits random walking at the beginning, with the accuracy of prediction on phase feedback direction, to allow the neural-network to step into the training range for fast convergence. This neural-network-based iterative method of phase detection works tens of times faster than the commonly used stochastic parallel gradient descent approach (SPGD) using a single-detector and random dither when both are tested with random phase perturbations.

Wang, Dan↗

Massively Multi-core Acceleration of a Document-Similarity Classifier to Detect Web Attacks

This paper describes our approach to adapting a text document similarity classifier based on the Term Frequency Inverse Document Frequency (TFIDF) metric to two massively multi-core hardware platforms. The TFIDF classifier is used to detect web attacks in HTTP data. In our parallel hardware approaches, we design streaming, real time classifiers by simplifying the sequential algorithm and manipulating the classifier's model to allow decision information to be represented compactly. Parallel implementations on the Tilera 64-core System on Chip and the Xilinx Virtex 5-LX FPGA are presented. For the Tilera, we employ a reduced state machine to recognize dictionary terms without requiring explicit tokenization, and achieve throughput of 37MB/s at slightly reduced accuracy. For the FPGA, we have developed a set of software tools to help automate the process of converting training data to synthesizable hardware and to provide a means of trading off between accuracy and resource utilization. The Xilinx Virtex 5-LX implementation requires 0.2% of the memory used by the original algorithm. At 166MB/s (80X the software) the hardware implementation is able to achieve Gigabit network throughput at the same accuracy as the original algorithm.

97 MATHEMATICS AND COMPUTING↗

Entity Resolution at Large Scale: Benchmarking and Algorithmics

We seek scalable benchmarks for entity resolution problems. Solutions to these problems range from trivial approaches such as string sorting to sophisticated methods such as statistical relational learning. The theoretical and practical complexity of these approaches varies widely, so one of the primary purposes of a benchmark will be to quantify the trade-off between solution quality and runtime. We are motivated by the ubiquitous nature of entity resolution as a fundamental problem faced by any organization that ingests large amounts of noisy text data. A benchmark is typically a rigid specification that provides an objective measure usable for ranking implementations of an algorithm. For example the Top500 and HPCG500 bench- marks rank supercomputers based on their performance of dense and sparse linear algebra problems (respectively). These two benchmarks require participants to report FLOPS counts attainable on various machines. Our purpose is slightly different. Whereas the supercomputing benchmarks mentioned above hold algorithms constant and aim to rank machines, we are primarily interested in ranking algorithms. As mentioned above, entity resolution problems can be approached in completely different ways. We believe that users of our benchmarks must decide what sort of procedure to run before comparing implementations and architectures. Eventually, we also wish to provide a mechanism for ranking machines while holding algorithmic approach constant . Our primary contributions are parallel algorithms for computing solution quality mea- sures per entity. We find in some real datasets that many entities are quite easy to resolve while others are difficult, with a heavy skew toward the former case. Therefore, measures such as global confusion matrices, F measures, etc. do not meet our benchmarking needs. We design methods for computing solution quality at the granularity of a single entity in order to know when proposed solutions do well in difficult situations (perhaps justifying extra computational), or struggling in easy situations. We report on progress toward a viable benchmark for comparing entity resolution algo- rithms. Our work is incomplete, but we have designed and prototyped several algorithms to help evalute the solution quality of competing approaches to these problems. We envision a benchmark in which the objective measure is a ratio of solution quality to runtime.

97 MATHEMATICS AND COMPUTING↗

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↗

Discovering causal structure with reproducing-kernel Hilbert space ε -machines

We merge computational mechanics’ definition of causal states (predictively equivalent histories) with reproducing-kernel Hilbert space (RKHS) representation inference. The result is a widely applicable method that infers causal structure directly from observations of a system’s behaviors whether they are over discrete or continuous events or time. A structural representation—a finite- or infinite-state kernel ϵ-machine—is extracted by a reduced-dimension transform that gives an efficient representation of causal states and their topology. In this way, the system dynamics are represented by a stochastic (ordinary or partial) differential equation that acts on causal states. We introduce an algorithm to estimate the associated evolution operator. Paralleling the Fokker–Planck equation, it efficiently evolves causal-state distributions and makes predictions in the original data space via an RKHS functional mapping. We demonstrate these techniques, together with their predictive abilities, on discrete-time, discrete-value infinite Markov-order processes generated by finite-state hidden Markov models with (i) finite or (ii) uncountably infinite causal states and (iii) continuous-time, continuous-value processes generated by thermally driven chaotic flows. The method robustly estimates causal structure in the presence of varying external and measurement noise levels and for very high-dimensional data.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Phasor-Measurement-Unit-Based Data Analytics Using Digital Twin and PhasorAnalytics Software

A major objective of this project was to apply GE’s commercial machine learning and data analytics toolsets to large-scale, real-world, anonymized Phasor Measurement Unit (PMU) datasets in order to extract signatures, correlated and/or causal factors, and precursor patterns associated with significant power system phenomena. The project had a particular emphasis on extraction of insights relevant to asset health monitoring, real-time load modeling and cybersecurity monitoring. Additionally, the team was directed to undertake a comprehensive data quality analysis for the provided datasets and encouraged to estimate the ‘machine-learning readiness’ of the datasets by documenting any major obstacles to the application of commercial machine learning algorithms. To accomplish the aforementioned objectives, the project team’s work centered around the identification of key event signatures and application of the identified event signatures for event detection and event classification. The industry-validated, semi-supervised machine learning strategy employed for event signature identification involved several major tasks, including data-preprocessing, generation of an overabundance of features, normal data identification, normality modeling, and event signature identification through a methodical, quantitative ranking of features in order of relevance to each studied event type. Throughout the project, data quality issues and mitigation techniques were investigated. In this report, insights are provided regarding the readiness of the provided synchrophasor datasets for application of machine learning and data analytics. The methodologies employed for this technical strategy are summarized in this report. With regards to data preprocessing and feature generation, the provided Training and Test Datasets were ingested into GE’s big data environment. Subsequently, the team applied bad data cleansing and data imputation scripts, event detection scripts, and application programming interfaces (APIs) to the datasets for convenient data access. The project team completed development and validation of dozens of physics-based, statistics-based and transformation-based feature functions used for the extraction of over 60 synchrophasor features. Using a new parallel feature generation technology developed on this project, over 60 features have been rapidly generated for the full two years’ worth of Training and Test Dataset data associated with both the Eastern and Western interconnects. Even accommodating for temporal down-sampling inherent to the feature extraction procedure, this parallel feature generation activity resulted in a massive feature set with a storage requirement approximately equal to that of the raw training dataset itself. With regards to normal data identification and normality modeling, a normality model was built using the feature data extracted from the Training Dataset and iteratively refined subsequent to incremental adjustments and expansions of the Training Dataset feature data. With respect to event characterization and signature identification, an event signature identification pipeline was developed and used in conjunction with the normality model to identify over 15 event signatures for key event categories within the Training Dataset. The identified event signatures were used to characterize hundreds of key events in terms of relative severity, duration, and location of the event. An investigation was undertaken to identify correlated and causal factors involved in transformer events. A separate investigation into temporal trends in ring-down analysis results was undertaken to determine possible associations between system dynamics and various other factors such as loading, season or year. To validate the identified event signatures, additional work was undertaken to develop signature-based anomaly detection and classification tools suitable for convenient application to the synchrophasor datasets. The anomaly detection and classification tools, suitable for online application, were then applied to the entirety of the Eastern Interconnect Training and Test Datasets. Performance of the event detection and classification tools was evaluated upon receipt of the Test Dataset event logs (i.e., the labels for events contained in the Test Dataset), and promising results were obtained despite several challenges (documented herein) associated with application of supervised or semi-supervised machine learning methods to large-scale, anonymized datasets. Finally, the detection and classification tools were used to detect, classify, and characterize thousands of new events not included in the original event logs provided by the DOE within both the Training and Test Datasets.

24 POWER TRANSMISSION AND DISTRIBUTION↗

A preliminary user-friendly, digital console for the control room parameters supervision in old-generation Nuclear Plants

Improvements in the awareness of a system status is an essential requirement to achieve safety in every kind of plant. In particular, in the case of Nuclear Power Plants (NPPs), a progress is crucial to enhance the Human Machine Interface (HMI) in order to optimize monitoring and analyzing processes of NPP operational states. Firstly, as old-fashioned plants are concerned, an upgrading of the whole console instrumentation is desirable in order to replace an analog visualization with a full-digital system. In this work, we present a novel instrument able to interface the control console of a nuclear reactor, developed by using CompactRio, a National Instruments embedded architecture and its dedicated programming language. This real-time industrial controller composed by a real-time processor and FPGA modules has been programmed to visualize the parameters coming from the reactor, and to storage and reproduce significant conditions anytime. This choice has been made on the basis of the FPGA properties: high reliability, determinism, true parallelism and re-configurability, achieved by a simple programming method, based on LabVIEW real-time environment. The system architecture exploits the FPGA capabilities of implementing custom timing and triggering, hardware-based analysis and co-processing, and highest performance control algorithms. Data stored during the supervisory phase can be reproduced by loading data from a measurement file, re-enacting worthwhile operations or conditions. The system has been thought to be used in three different modes, namely Log File Mode, Supervisory Mode and Simulation Mode. The proposed system can be considered as a first step to develop a more complete Decision Support System (DSS): indeed this work is part of a wider project that includes the elaboration of intelligent agents and meta-theory approaches. A synoptic has been created to monitor every kind of action on the plant through an intuitive sight. Furthermore, another important aim of this work is the possibility to have a front panel available on a web interface: CompactRio acts as a remote server and it is accessible on a dedicated LAN. This supervisory system has been tested and validated on the basis of the real control console for the 1-MW TRIGA reactor RC-1 at the ENEA, Casaccia Research Center. In this paper we show some results obtained by recording each variable as the reactor reaches its maximum level of power. The choice of a research reactor for testing the developed system relies on its training and didactic importance for the education of plant operators: in this context a digital instrument can offer a better user-friendly tool for learning and training. It is worthwhile to remark that such a system does not interfere with the console instrumentation, the latter continuing to preserve the total control. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Two-dimensional materials for bio-realistic neuronal computing networks

Two-dimensional (2D) van der Waals materials have found broad utility in a diverse range of applications including electronics, optoelectronics, renewable energy, and quantum information technologies. Meanwhile, exponentially growing digital data coupled with the ubiquity of artificial intelligence algorithms have generated significant interest in edge neuromorphic computing as an alternative to centralized cloud computing. The drive to incorporate neuroscience principles into computing hardware is motivated by the low power consumption, parallel processing, and reconfigurability of the human brain. The diverse library of 2D materials with atomic-level thicknesses, exceptional electrostatic tunability, and integration versatility is particularly well-suited for realizing bio-realistic synaptic and neuronal functionality. Here, we summarize past and present work in this field and outline the frontier challenges that have not yet been overcome. Here we also delineate potential solutions and suggest that the neuroscience principles of criticality and synchrony have the potential to inspire breakthrough applications of 2D materials in neuronal computing networks.

36 MATERIALS SCIENCE↗

Parallel physics-informed neural networks via domain decomposition

Here we develop a distributed framework for the physics-informed neural networks (PINNs) based on two recent extensions, namely conservative PINNs (cPINNs) and extended PINNs (XPINNs), which employ domain decomposition in space and in time-space, respectively. This domain decomposition endows cPINNs and XPINNs with several advantages over the vanilla PINNs, such as parallelization capacity, large representation capacity, efficient hyperparameter tuning, and is particularly effective for multi-scale and multi-physics problems. Here, we present a parallel algorithm for cPINNs and XPINNs constructed with a hybrid programming model described by MPI + X, where X ∈ {CPUs, GPUs}. The main advantage of cPINN and XPINN over the more classical data and model parallel approaches is the flexibility of optimizing all hyperparameters of each neural network separately in each subdomain. We compare the performance of distributed cPINNs and XPINNs for various forward problems, using both weak and strong scalings. Our results indicate that for space domain decomposition, cPINNs are more efficient in terms of communication cost but XPINNs provide greater flexibility as they can also handle time-domain decomposition for any differential equations, and can deal with any arbitrarily shaped complex subdomains. To this end, we also present an application of the parallel XPINN method for solving an inverse diffusion problem with variable conductivity on the United States map, using ten regions as subdomains.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Scalable Hyper-parameter Estimation for Gaussian Process Based Time Series Analysis

Gaussian process (GP) is increasingly becoming popular as a kernel machine learning tool for non-parametric data analysis. Recently, GP has been applied to model non-linear dependencies in time series data. GP based analysis can be used to solve problems of time series prediction, forecasting, missing data imputation, change point detection, anomaly detection, etc. But the use of GP to handle massive scientific time series data sets has been limited, owing to its expensive computational complexity. The primary bottleneck is the handling of the covariance matrix whose size is quadratic in the length of the time series. In this paper we propose a scalable method that exploit the special structure of the covariance matrix for hyper-parameter estimation in GP based learning. The proposed method allows estimation of hyper parameters associated with GP in quadratic time, which is an order of magnitude improvement over standard methods with cubic complexity. Moreover, the proposed method does not require explicit computation of the covariance matrix and hence has memory requirement linear to the length of the time series as opposed to the quadratic memory requirement of standard methods. To further improve the computational complexity of the proposed method, we provide a parallel version to concurrently estimate the log likelihood for a set of time series which is the key step in the hyper-parameter estimation. Performance results on a multi-core system show that our proposed method provides significant speedups as high as 1000, even when running in serial mode, while maintaining a small memory footprint. The parallel version exploits the natural parallelization potential of the serial algorithm and is shown to perform significantly better than the serial faster algorithm, with speedups as high as 10.

Chandola, Varun↗

A Parametric, Data-Driven, Non-Intrusive Reduced-Order Model Framework for Crystal Plasticity Simulations of Voids

The influence of the internal structure at micrometer length scales on the deformation of polycrystalline materials can be effectively captured using crystal plasticity finite element methods (CPFEM). However, the complexity and nonlinearity of the deformation equations CPFEM solves demand significant computational power and resources to achieve accurate predictions, limiting its broader application. To address this challenge, we have identified a reduced-order representation of the complex data in order to establish a computationally efficient reduced-order models (ROM) and drastically reduce the computational expense of CPFEM. Specifically, in this work, we developed a parametric, data-driven, and non-intrusive ROM framework for CPFEM using proper orthogonal decomposition (POD) and sparse variational Gaussian process (SVGP) regression for single-crystal microstructures under tensile loading conditions. The developed protocol enables one to compress field into a latent/low-dimensional space described by principal component analysis (PCA) via the singular value decomposition (SVD) algorithm. As a result, the high-dimensional data are reduced to a significantly smaller amount of dimensions with POD bases and POD coefficients. Furthermore, we deployed an ensemble of SVGPs—extended from the classical Gaussian process (GP) regression for scalability and handling big data—in a massively parallel manner to train and predict latent POD coefficients using known POD bases from a set of previously obtained simulations results. Lastly, using the predicted POD coefficients, we reconstructed the full-field results and showed reasonable agreement compared with the true values obtained from running CPFEM. The developed framework is validated with a set of CPFEM simulations of a single embedded void in single-crystal aluminum alloy. While the framework is broadly applicable, this work specifically focuses on single-crystal microstructures, a single load case (e.g., tensile), and a specific void geometry (spherical).

Anisotropy↗

Towards a more general understanding of the algorithmic utility of recurrent connections

Lateral and recurrent connections are ubiquitous in biological neural circuits. Yet while the strong computational abilities of feedforward networks have been extensively studied, our understanding of the role and advantages of recurrent computations that might explain their prevalence remains an important open challenge. Foundational studies by Minsky and Roelfsema argued that computations that require propagation of global information for local computation to take place would particularly benefit from the sequential, parallel nature of processing in recurrent networks. Such “tag propagation” algorithms perform repeated, local propagation of information and were originally introduced in the context of detecting connectedness, a task that is challenging for feedforward networks. Here, we advance the understanding of the utility of lateral and recurrent computation by first performing a large-scale empirical study of neural architectures for the computation of connectedness to explore feedforward solutions more fully and establish robustly the importance of recurrent architectures. In addition, we highlight a tradeoff between computation time and performance and construct hybrid feedforward/recurrent models that perform well even in the presence of varying computational time limitations. We then generalize tag propagation architectures to propagating multiple interacting tags and demonstrate that these are efficient computational substrates for more general computations of connectedness by introducing and solving an abstracted biologically inspired decision-making task. Our work thus clarifies and expands the set of computational tasks that can be solved efficiently by recurrent computation, yielding hypotheses for structure in population activity that may be present in such tasks.

59 BASIC BIOLOGICAL SCIENCES↗

Cybersecurity for Grid Connected eXtreme Fast Charging (XFC) Station (CyberX) (Final Scientific/Technical Report)

This report summarizes the activities conducted under the DOE VTO funded project DE- EE0008451, where ABB Inc. (ABB), in collaboration with Idaho National Laboratory (INL), APS Global (APS), and XOS Trucks (XOS) pursued the development of a cyber-resilient extreme fast charging (XFC) management system. This project entitled Cybersecurity for Grid Connected eXtreme Fast Charging (XFC) Station (CyberX) focuses on a resilient architecture for smart charging EV Supply Equipment (EVSE) device control and Coordinated Anomaly Detection System (CADS) features that can be added at the charging site depot level to increase cybersecurity. The project was split into two budget periods focused first on developing the threat model and resilient control concepts and second on testing, improving, and validating those developed resilient control algorithms and features with a focus on key vulnerabilities identified during the threat assessment portion of the project. During the first budget period of the CyberX project, the ABB led team focused on activities to identify, model, and quantitatively prioritize high-impact attack scenarios with potential cyber-physical effects while also modeling and developing concepts for a resilient control system that could securely address integration of DERs and other resources with EV charging. Development of the security focused XFC management system (XMS) was accomplished first by offline simulation using a developed XFC station or depot with 480V input level and simulating measurement inputs to monitoring and control systems in concept development. A representative distribution grid model was developed supporting an EV charging site model with BESS and 6 general EV charging models. These EV charging models allowed multiple configurations of charging level, multiple connected protection and measurement devices, and simulation function to show general compromise of EV, BESS, and protection features based on parallel threat analysis. During the second budget period, the EV site and supporting systems model was developed in more detail and converted from offline model to real-time to real-time with EV charging hardware in the loop (HIL). The resilient control architecture developed as concept in the first part of the project was further tested and validated for integration of local energy resources and XFC charging station site equipment while maintaining cybersecure operating principles. The proposed resilient architecture for smart charging and cybersecurity features consists of two main concepts developed and tested within the project. The first concept is an XFC management system (XMS) consisting of a hardware gateway, software platform, and Supervisory Control and Data Acquisition (SCADA) or Distribution Management System integration components. The second concept is a Coordinated Anomaly Detection System (CADS) which forms a primarily software-related subsystem of the total CyberX solution focused on monitoring system measurements, estimation of measurement states, and predicting current at the utility point of interaction based on machine learning for anomaly detection.

33 ADVANCED PROPULSION SYSTEMS↗

PIPER: Performance Insights for Programmers and Exascale Runtimes

During the lifetime of the project, the PIPER at the University of Utah worked in collaboration with the team from the partner institutions to establish a set of software tools for analyzing and improving the performance parallel computing tools. These activities complement and complete the work performed by the other X-Stack 2 projects with the definition of a common data models and computing infrastructures, which have been a central element in the targeted performance analysis pipeline and new programming models. The University of Utah team have focused mainly on developing data analysis and visualization systems to facilitate exploration, comparison and evaluation of the behavior and performance of complex algorithms in modern HPC infrastructures. This is involved several demonstrations of the performance achieved and presentation of the results in a way that facilitates the users understand bottlenecks and ways to overcome them. One major focus of the University of Utah activities has been the analysis of the performance characteristics of Fat-Tree and Dragonfly topology networks. The work involved the development of practical visualization systems especially in collaboration with the team at LLNL, and published several papers and gave invited talks at prestigious venues such as the Salishan Conference on High-Speed Computing or the International Conference on High Performance Computing, Data, and Analytics. In addition, the university of Utah team has advanced the core activities on analyzing the performance characteristics of I/O and other data movement systems and their practical use for scaling Uintah and other simulation codes poised for use in the exascale computing environments of the future.

42 ENGINEERING↗

Scalable training of trustworthy and energy-efficient predictive graph foundation models for atomistic materials modeling: a case study with HydraGNN

We present our work on developing and training scalable, trustworthy, and energy-efficient predictive graph foundation models (GFMs) using HydraGNN, a multi-headed graph convolutional neural network architecture. HydraGNN expands the boundaries of graph neural network (GNN) computations in both training scale and data diversity. It abstracts over message passing algorithms, allowing both reproduction of and comparison across algorithmic innovations that define nearest-neighbor convolution in GNNs. This work discusses a series of optimizations that have allowed scaling up the GFMs training to tens of thousands of GPUs on datasets consisting of hundreds of millions of graphs. Our GFMs use multitask learning (MTL) to simultaneously learn graph-level and node-level properties of atomistic structures, such as energy and atomic forces. Using over 154 million atomistic structures for training, we illustrate the performance of our approach along with the lessons learned on two state-of-the-art US Department of Energy (US-DOE) supercomputers, namely the Perlmutter petascale system at the National Energy Research Scientific Computing Center and the Frontier exascale system at Oak Ridge Leadership Computing Facility. The HydraGNN architecture enables the GFM to achieve near-linear strong scaling performance using more than 2000 GPUs on Perlmutter and 16,000 GPUs on Frontier.

97 MATHEMATICS AND COMPUTING↗

Considerations regarding the Use of Computer Vision Machine Learning in Safety-Related or Risk-Significant Applications in Nuclear Power Plants

With the advancements made to date in the field of artificial intelligence (AI), significant potential exists to utilize AI capabilities for nuclear power plant (NPP) applications. AI can replicate human decision making and it is usually faster and more accurate than humans. For implementations that impact critical NPP applications (e.g., safety-related or non-safety systems that potentially affect overall plant risk), a deeper safety analysis of the AI methods is necessary. AI applied to NPP operations could resemble the use of digital I&C (DI&C) because such applications involve digital computer hardware and custom-designed software that input plant data, execute complex software algorithms, and output the results to a system or licensed human operator to potentially provoke an action. For AI methods to be compliant with current safety requirements for DI&C, AI compatibility must be evaluated, and AI-related gaps may exist that prevent the prompt deployment of AI in NPPs. This effort aims to evaluate how example AI technologies align with the DI&C safety framework, and discusses how they could be analyzed, modeled, tested, and validated in a manner similar to typical DI&C technologies. Because AI is a broad field that encompasses areas such as machine learning (ML), natural language processing, and computer vision, this research focused on a subset of methods categorized as the computer vision ML (CVML) methods. This report explores two CVML use cases, gauge reading and fire watch, considered relevant to the DI&C standards, as they could play a safety-critical role. For the gauge reading use case, a CVML-enabled technology that can read gauges at oblique angles is utilized. For the fire watch use case, a CVML-enabled technology is utilized that migrates fire watch from a manual (human) approach to automated fire detection. These use cases are mainly intended to give context to the CVML system discussion. This effort assumes the worst-case scenario, with the CVML system being used to replace a safety-related or risk-significant system, thus requiring evaluation. Evaluating CVML against most of the relevant safety requirements for DI&C yielded several CVML-specific considerations due to the uniqueness of its characteristics in comparison with typical DI&C systems. For example, CVML models often employ commonly used (open-source) datasets, and it is not always possible to determine the level of overlap among open-source datasets. Therefore, the independence of the developed CVML models when demonstrating diversity is questionable, therefore creating vulnerability to common cause failure (CCF). The design verification process is also impacted since the data overlap could result in overestimation of the software validation and verification (V&V) performance results. Section 2 of this report evaluates a list of the identified CVML-specific characteristics and discusses the resulting considerations and potential solutions in the context of each referenced requirement. A summation is provided in Section 3. This report is not to be used as a guideline. It was developed to identify and consider issues in the implementation of ML technologies used to augment activities that may have a bearing on plant operation. The report draws parallels to the use of DI&C technologies, for which many standards are available to guide their use in nuclear plant operation. It considers the technologies and some of the potential implications of their use in safety-related applications but is not intended to address regulatory or licensing related issues.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)

The development of modern processors exhibits two trends that complicate the optimizations of modern software. The first is the increasing sensitivity of processors' throughput to irregularities in computation. With more processors produced through a massive integration of simple cores, future systems will increasingly favor regular data-level parallel computations, but deviate from the needs of applications with complex patterns. Some evidences are already shown on Graphic Processing Units (GPU): Irregular data accesses (e.g., indirect references A[D[i]]) and conditional branches are limiting many GPU applications' performance at a level an order of magnitude lower than the peak of GPU. The second hardware trend is the growing gap between memory bandwidth and the aggregate speed—that is, the sum of all cores' computing power—of a Chip Multiprocessor (CMP). Despite the capped growth of the peak CPU speed, the aggregate speed of a CMP keeps increasing as more cores get into a single chip. It is expected that by 2018, node concurrency in an exascale system will increase by hundreds of times, whereas, memory bandwidth will expand by only 10 to 20 times. Consequently, data movement and storage is expected to consume more than 70% of the total system power. Bridging this gap is difficult; the complexities of modern CMP memory hierarchy make it even harder: Data cache becomes shared among computing units, and the sharing is often non-uniform—whether two computing units share a cache depends on their proximity and the level of the cache. On the recent IBM Power7 architecture, for instance, four hardware contexts (or SMT threads) in a core share the entire memory hierarchy, all cores in one chip share an on-chip L3 cache, and cores across chips share L3 and main memory through off-chip connections. These two trends complicate the translation of computing power into performance, especially for a program with either intensive data accesses or complex patterns in data accesses or control flow paths. Unfortunately, both attributes present and will persist in a class of important applications. For instance, many scientific simulations deal with a large volume of data. And meanwhile, as most real-world processes are non-uniform and evolving (e.g., the evolution of a galaxy or the process of a drug injection), both the computations and data accesses of these programs tend to be irregular and dynamically changing. Currently, the lack of support to these applications on modern CMP severely limits their performance. On GPU, as our recent study shows and other studies echo, performance enhancement of a factor of integers is possible when memory accesses or control flows are streamlined for a set of GPU applications. On multicore CPU, our studies show that traditional locality enhancement, for being oblivious to the new features of multicore memory hierarchy, may even cause large slowdown to data-intensive dynamic applications. The severity of the issues is expected to worsen as the two hardware trends continue. Some recent studies try to match software with the trends, but in a limited scope or manner. For irregularities on GPU, most studies focus on irregularities analyzable through static analysis (e.g., data accesses in regular loops). Dynamic irregularities are harder to address because the needed analysis and transformations typically have to happen at run time. Some other research resorts to hardware extensions, an actual adoption of which is unclear for the entailed space cost and complexity. For data locality, recent years have seen some exploitations of the new memory hierarchy on multicore for performance, but most of them are on process or thread scheduling, rather than program transformations. Our recent study reveals that program-level transformations may magnify the scheduling benefits by a factor of seven, concluding that program-level transformation should play a central role for data locality enhancement on modern CMP. But research in this direction has been sparse, and most have focused on data layout or cache performance modeling, rather than program transformations to match with the new memory hierarchy features. Overall, it is still an open question how to bridge the gap between dynamic computations and the two prominent properties of modern processors. The goal of this project is to develop a set of techniques and software tools to enhance the matching between memory accesses in dynamic simulations and the prominent features of modern and future CMP, alleviating the memory performance issues for petascale and exascale computing. This report summarizes the discoveries and products produced throughout this project. It includes free launch, a new software approach to overcoming the shortcomings of both methods; coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution; algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Supporting dynamic parallelism is important for GPU to benefit a broad range of applications. There are currently two fundamental ways for programs to exploit dynamic parallelism on GPU: a software-based approach with software-managed worklists, and a hardware-based approach through dynamic subkernel launches. Neither is satisfactory. The former is complicated to program and is often subject to some load imbalance; the latter suffers large runtime overhead. In this work, we propose free launch, a new software approach to overcoming the shortcomings of both methods. It allows programmers to use subkernel launches to express dynamic parallelism. It employs a novel compiler-based code transformation named subkernel launch removal to replace the subkernel launches with the reuse of parent threads. Coupled with an adaptive task assignment mechanism, the transformation reassigns the tasks in the subkernels to the parent threads with a good load balance. The technique requires no hardware extensions, immediately deployable on existing GPUs. It keeps the programming convenience of the subkernel launch-based approach while avoiding its large runtime overhead. Meanwhile, its superior load balancing makes it outperform manual worklist-based techniques by 3X on average. The work was published at the 48th Annual IEEE/ACM International Symposium on Microarchitecture (Micro'2015). A Graphic Processing Unit (GPU) system is typically equipped with many types of memory (e.g., global, constant, texture, shared, cache). Data placement determines what data are placed on which type of memory, essential for GPU memory performance. Prior optimizations of data placement always require a single view of a data object on memory, which limits the optimization effectiveness. In this work, we propose coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution. We demonstrate that under certain conditions, the multiple views can remain incoherent while facilitating enhanced data placement. We present a theorem and some compiler support to ensure the soundness of the usage of coherence-free multiview. We further develop reference-discerning data placement, a new way to enhance data placements on GPU. It enables more flexible data placements by using coherence-free multiview to leverage the slack in coherence requirement of some GPU programs. Experiments on three types of GPU systems show that, with less than 200KB space cost, the new data placement technique can provide a 1.6X average (up to 4.27X) speedup. In addition, we have examined algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Computing distances among data points is an essential part of many important algorithms in data analytics, graph analysis, and other domains. In each of these domains, developers have spent significant manual effort optimizing algorithms, often through novel applications of the triangle equality, in order to minimize the number of distance computations in the algorithms. In this work, we observe that many algorithms across these domains can be generalized as an instance of a generic distance-related abstraction. Based on this abstraction, we derive seven principles for correctly applying the triangular inequality to optimize distance-related algorithms. Guided by the findings, we develop Triangular OPtimizer (TOP), the first software framework that is able to automatically produce optimized algorithms that either matches or outperforms manually designed algorithms for solving distance-related problems. TOP achieves up to 237x speedups and 2.5X on average. The work has been published at the 32nd International Conference on Machine Learning (ICML'15) and the 41st International Conference on Very Large Data Bases (VLDB'15).

42 ENGINEERING↗