Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “algorithm selection”

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 379 records · Page 21

Metric Learning to Accelerate Convergence of Operator Splitting Methods

Recent developments in machine learning have led to promising advances in accelerating the solution of constrained optimization problems. Increasing demand for real-time decision-making capabilities in applications such as artificial intelligence and optimal control has led to a variety of proposed strategies for learning to produce fast solutions to optimization problems. For example, recent works have shown that it is possible to accelerate the convergence of optimization algorithms by learning to select their parameters, such as gradient descent stepsizes. This work proposes a new approach, in which the underlying metric spaces of proximal operator splitting algorithms are learned to maximize convergence rate. While prior works in optimization theory have derived optimal metrics in simple cases, no such result exists for many practical problem forms including general Quadratic Programming (QP). This paper shows how differentiable optimization can enable the end-to-end learning of proximal metrics, enhancing the convergence of proximal algorithms for QP problems beyond what is possible based on known theory. Additionally, the results illustrate a strong connection between the learned proximal metrics and active constraints at the optima, leading to an interpretation in which the predicted proximal metrics can be viewed as a form of active set prediction.

King, Ethan [BATTELLE (PACIFIC NW LAB)]↗

Benchmarking for AI for Science

AI has been instrumental for recent developments in a number of domains of the sciences. With several hundred machine learning (ML) algorithms and models, and numerous AI-specific hardware platforms, a common quest for all scientists working on AI for Science is around the selection of machine learning algorithm(s) to solve their domain-specific scientific problems. A number of different initiatives around AI Benchmarking have been set up and have been useful in understanding the benefits of different ML algorithms for different tasks.However, with the majority of these AI Benchmarking initiatives focusing on the conventional notions of benchmarking, where the focus is purely runtime performance (such as training time or inference time), their suitability for benchmarking different ML algorithms for solving scientific problems has been viewed as a performance problem even though both are hardly the same. To make reasonable, explainable, and justifiable advancements in science using AI, it is critical to focus on the merits of these algorithms in handling different domain science problems. In other words, more emphasis must be given on Benchmarking for AI for Science than AI Benchmarking. The vision of the former is not only to assess the performance of ML algorithms, but also to assess, and understand the benefits and merits of different ML algorithms in handling scientific problems. Benchmarking for AI for Science, instead of pure performance focused AI Benchmarking, has several benefits: (i) it has the potential to offer advances in the sciences, much more rapidly than through pure performance-based AI methods, (ii) it will encourage the community to focus on developing better domain-specific AI techniques, particularly given the provision for being able to benchmark different techniques, and (iii) it will encourage hardware manufacturers to focus on developing science-specific hardware subsystems.

Thiyagalingam, Jeyan↗

X-ray scaling relations from a complete sample of the richest maxBCG clusters

Here, we use a complete sample of 38 richest maxBCG clusters to study the ICM–galaxy scaling relations and the halo mass selection properties of the maxBCG algorithm, based on X-ray and optical observations. The clusters are selected from the two largest bins of optical richness in the Planck stacking work with the maxBCG richness N 200 ≥ 78. We analyse their Chandra and XMM–Newton data to derive the X-ray properties of the ICM. We then use the distribution of P(X|N), X = T X , L X , Y X , to study the mass selection P(M|N) of maxBCG. Compared with previous works based on the whole richness sample, a significant fraction of blended systems with boosted richness is skewed into this richest sample. Parts of the blended haloes are picked apart by the redMaPPer, an updated red-sequence cluster finding algorithm with lower mass scatter. Moreover, all the optical blended haloes are resolved as individual X-ray haloes, following the established L X -T X and L X -Y X relations. We further discuss that the discrepancy between ICM–galaxy scaling relations, especially for future blind stacking, can come from several factors, including miscentring, projection, contamination of low-mass systems, mass bias, and covariance bias. We also evaluate the fractions of relaxed and cool core clusters in our sample. Both are smaller than those from SZ or X-ray selected samples. Moreover, disturbed clusters show a higher level of mass bias than relaxed clusters.

79 ASTRONOMY AND ASTROPHYSICS↗

SU-E-T-348: Verification MU Calculation for Conformal Radiotherapy with Multileaf Collimator Using Report AAPM TG 114

Purpose: The purpose of this work was to develop and validate an open source independent MU dose calculation software for 3D conformal radiotherapy with multileaf high and low resolution according to the report of AAPM TG 11 Methods: Treatment plans were done using Iplan v4.5 BrainLAB TPS. A 6MV photon beam produced by Primus and Novalis linear accelerators equipped with an Optifocus MLC and HDMLC, respectively. TPS dose calculation algorithms were pencil beam and Monte Carlo. 1082 treatments plans were selected for the study. The algorithm was written in free and open source CodeBlocks C++ platform. Treatment plans were imported by the software using RTP format. Equivalent size field is obtained from the positions of the leaves; the effective depth of calculation can be introduced by TPS's dosimetry report or automatically calculated starting from SSD. The inverse square law is calculated by the 3D coordinates of the isocenter and normalization point of the treatment plan. The dosimetric parameters TPR, Sc, Sp and WF are linearly interpolated. Results: 1082 plans of both machines were analyzed. The average uncertainty between the TPS and the independent calculation was −0.43% ± 2.42% [−7.90%, 7.50%]. Specifically for the Primus the variation obtained was −0.85% ± 2.53% and for the Novalis 0.00% ± 2.23%. Data show that 94.8% of the cases the uncertainty was less than or equal to 5%, while 98.9% is less than or equal to 6%. Conclusion: The developed software is appropriate for use in calculation of UM. This software can be obtained upon request.

07 ISOTOPE AND RADIATION SOURCES↗

Gradient-Free Construction of Active Subspaces for Dimension Reduction in Complex Models with Applications to Neutronics

Recent developments in the field of reduced-order modeling---and, in particular, active subspace construction---have made it possible to efficiently approximate complex models by constructing low-order response surfaces based upon a small subspace of the original high-dimensional parameter space. These methods rely upon the fact that the response tends to vary more prominently in a few dominant directions defined by linear combinations of the original inputs, allowing for a rotation of the coordinate axis and a consequent transformation of the parameters. In this work, we discuss a gradient-free active subspace algorithm that is feasible for high-dimensional parameter spaces where finite-difference techniques are impractical. This analysis extends the gradient-free algorithm introduced in [A. Lewis, R. Smith, and B. Williams, Comput. Math. Appl., 72 (2016), pp. 1603--1615] in two significant ways: (i) we introduce an initialization algorithm to identify lower-dimensional subspaces of influential directions to seed the gradient-free algorithm for high-dimensional problems, and (ii) we analyze dimension selection criteria to verify the algorithms. Finally, we illustrate the initialized gradient-free active subspace algorithm for a neutronics example implemented with SCALE6.1 for input dimensions up to 7700.

97 MATHEMATICS AND COMPUTING↗

Sensor placement algorithm development to maximize the efficiency of acid gas removal unit for integrated gasifiction combined sycle (IGCC) power plant with CO2 capture

Future integrated gasification combined cycle (IGCC) power plants with CO{sub 2} capture will face stricter operational and environmental constraints. Accurate values of relevant states/outputs/disturbances are needed to satisfy these constraints and to maximize the operational efficiency. Unfortunately, a number of these process variables cannot be measured while a number of them can be measured, but have low precision, reliability, or signal-to-noise ratio. In this work, a sensor placement (SP) algorithm is developed for optimal selection of sensor location, number, and type that can maximize the plant efficiency and result in a desired precision of the relevant measured/unmeasured states. In this work, an SP algorithm is developed for an selective, dual-stage Selexol-based acid gas removal (AGR) unit for an IGCC plant with pre-combustion CO{sub 2} capture. A comprehensive nonlinear dynamic model of the AGR unit is developed in Aspen Plus Dynamics® (APD) and used to generate a linear state-space model that is used in the SP algorithm. The SP algorithm is developed with the assumption that an optimal Kalman filter will be implemented in the plant for state and disturbance estimation. The algorithm is developed assuming steady-state Kalman filtering and steady-state operation of the plant. The control system is considered to operate based on the estimated states and thereby, captures the effects of the SP algorithm on the overall plant efficiency. The optimization problem is solved by Genetic Algorithm (GA) considering both linear and nonlinear equality and inequality constraints. Due to the very large number of candidate sets available for sensor placement and because of the long time that it takes to solve the constrained optimization problem that includes more than 1000 states, solution of this problem is computationally expensive. For reducing the computation time, parallel computing is performed using the Distributed Computing Server (DCS®) and the Parallel Computing® toolbox from Mathworks®. In this presentation, we will share our experience in setting up parallel computing using GA in the MATLAB® environment and present the overall approach for achieving higher computational efficiency in this framework.

97 MATHEMATICS AND COMPUTING↗

Sensor placement algorithm development to maximize the efficiency of acid gas removal unit for integrated gasification combined cycle (IGCC) power plant with CO{sub 2} capture

Future integrated gasification combined cycle (IGCC) power plants with CO{sub 2} capture will face stricter operational and environmental constraints. Accurate values of relevant states/outputs/disturbances are needed to satisfy these constraints and to maximize the operational efficiency. Unfortunately, a number of these process variables cannot be measured while a number of them can be measured, but have low precision, reliability, or signal-to-noise ratio. In this work, a sensor placement (SP) algorithm is developed for optimal selection of sensor location, number, and type that can maximize the plant efficiency and result in a desired precision of the relevant measured/unmeasured states. In this work, an SP algorithm is developed for an selective, dual-stage Selexol-based acid gas removal (AGR) unit for an IGCC plant with pre-combustion CO{sub 2} capture. A comprehensive nonlinear dynamic model of the AGR unit is developed in Aspen Plus Dynamics® (APD) and used to generate a linear state-space model that is used in the SP algorithm. The SP algorithm is developed with the assumption that an optimal Kalman filter will be implemented in the plant for state and disturbance estimation. The algorithm is developed assuming steady-state Kalman filtering and steady-state operation of the plant. The control system is considered to operate based on the estimated states and thereby, captures the effects of the SP algorithm on the overall plant efficiency. The optimization problem is solved by Genetic Algorithm (GA) considering both linear and nonlinear equality and inequality constraints. Due to the very large number of candidate sets available for sensor placement and because of the long time that it takes to solve the constrained optimization problem that includes more than 1000 states, solution of this problem is computationally expensive. For reducing the computation time, parallel computing is performed using the Distributed Computing Server (DCS®) and the Parallel Computing® toolbox from Mathworks®. In this presentation, we will share our experience in setting up parallel computing using GA in the MATLAB® environment and present the overall approach for achieving higher computational efficiency in this framework.

20 FOSSIL-FUELED POWER PLANTS↗

TH-EF-BRB-05: 4pi Non-Coplanar IMRT Beam Angle Selection by Convex Optimization with Group Sparsity Penalty

Purpose: Integrated beam orientation and fluence map optimization is expected to be the foundation of robust automated planning but existing heuristic methods do not promise global optimality. We aim to develop a new method for beam angle selection in 4π non-coplanar IMRT systems based on solving (globally) a single convex optimization problem, and to demonstrate the effectiveness of the method by comparison with a state of the art column generation method for 4π beam angle selection. Methods: The beam angle selection problem is formulated as a large scale convex fluence map optimization problem with an additional group sparsity term that encourages most candidate beams to be inactive. The optimization problem is solved using an accelerated first-order method, the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA). The beam angle selection and fluence map optimization algorithm is used to create non-coplanar 4π treatment plans for several cases (including head and neck, lung, and prostate cases) and the resulting treatment plans are compared with 4π treatment plans created using the column generation algorithm. Results: In our experiments the treatment plans created using the group sparsity method meet or exceed the dosimetric quality of plans created using the column generation algorithm, which was shown superior to clinical plans. Moreover, the group sparsity approach converges in about 3 minutes in these cases, as compared with runtimes of a few hours for the column generation method. Conclusion: This work demonstrates the first non-greedy approach to non-coplanar beam angle selection, based on convex optimization, for 4π IMRT systems. The method given here improves both treatment plan quality and runtime as compared with a state of the art column generation algorithm. When the group sparsity term is set to zero, we obtain an excellent method for fluence map optimization, useful when beam angles have already been selected. NIH R43CA183390, NIH R01CA188300, Varian Medical Systems; Part of this research took place while D. O’Connor was a summer intern at RefleXion Medical.

60 APPLIED LIFE SCIENCES↗

The CMS trigger system

This paper describes the CMS trigger system and its performance during Run 1 of the LHC. The trigger system consists of two levels designed to select events of potential physics interest from a GHz (MHz) interaction rate of proton-proton (heavy ion) collisions. The first level of the trigger is implemented in hardware, and selects events containing detector signals consistent with an electron, photon, muon, τ lepton, jet, or missing transverse energy. A programmable menu of up to 128 object-based algorithms is used to select events for subsequent processing. The trigger thresholds are adjusted to the LHC instantaneous luminosity during data taking in order to restrict the output rate to 100 kHz, the upper limit imposed by the CMS readout electronics. The second level, implemented in software, further refines the purity of the output stream, selecting an average rate of 400 Hz for offline event storage. The objectives, strategy and performance of the trigger system during the LHC Run 1 are described.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Feature selection with distance correlation

Choosing which properties of the data to use as input to multivariate decision algorithms—also known as feature selection—is an important step in solving any problem with machine learning. While there is a clear trend towards training sophisticated deep networks on large numbers of relatively unprocessed inputs (so-called automated feature engineering), for many tasks in physics, sets of theoretically well-motivated and well-understood features already exist. Working with such features can bring many benefits, including greater interpretability, reduced training and run time, and enhanced stability and robustness. We develop a new feature selection method based on distance correlation, and demonstrate its effectiveness on the tasks of boosted top- and W -tagging. Using our method to select features from a set of over 7,000 energy flow polynomials, we show that we can match the performance of much deeper architectures, by using only ten features and two orders-of-magnitude fewer model parameters. Published by the American Physical Society 2024

Astronomy & Astrophysics↗

Characterization of Pulmonary Nodules Based on Features of Margin Sharpness and Texture

Lung cancer is the leading cause of cancer-related deaths in the world, and one of its manifestations occurs with the appearance of pulmonary nodules. The classification of pulmonary nodules may be a complex task to specialists due to temporal, subjective, and qualitative aspects. Therefore, it is important to integrate computational tools to the early pulmonary nodule classification process, since they have the potential to characterize objectively and quantitatively the lesions. In this context, the goal of this work is to perform the classification of pulmonary nodules based on image features of texture and margin sharpness. Computed tomography scans were obtained from a publicly available image database. Texture attributes were extracted from a co-occurrence matrix obtained from the nodule volume. Margin sharpness attributes were extracted from perpendicular lines drawn over the borders on all nodule slices. Feature selection was performed by different algorithms. Classification was performed by several machine learning classifiers and assessed by the area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy. Highest classification performance was obtained by a random forest algorithm with all 48 extracted features. However, a decision tree using only two selected features obtained statistically equivalent performance on sensitivity and specificity.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Fast Demand Forecast of Electric Vehicle Charging Stations for Cell Phone Application

This paper describes the core cellphone application algorithm which has been implemented for the prediction of energy consumption at Electric Vehicle (EV) Charging Stations at UCLA. For this interactive user application, the total time of accessing database, processing the data and making the prediction, needs to be within a few seconds. We analyze four relatively fast Machine Learning based time series prediction algorithms for our prediction engine: Historical Average, kNearest Neighbor, Weighted k-Nearest Neighbor, and Lazy Learning. The Nearest Neighbor algorithm (k Nearest Neighbor with k=1) shows better performance and is selected to be the prediction algorithm implemented for the cellphone application. Two applications have been designed on top of the prediction algorithm: one predicts the expected available energy at the station and the other one predicts the expected charging finishing time. The total time, including accessing the database, data processing, and prediction is about one second for both applications.

Majidpour, Mostafa↗

Hashed Binary Search Sampling for Convolutional Network Training with Large Overhead Image Patches

Very large overhead imagery associated with ground truth maps has the potential to generate billions of training image patches for machine learning algorithms. However, random sampling selection criteria often leads to redundant and noisy-image patches for model training. With minimal research efforts behind this challenge, the current status spells missed opportunities to develop supervised learning algorithms that generalize over wide geographical scenes. In addition, much of the computational cycles for large scale machine learning are poorly spent crunching through noisy and redundant image patches. We demonstrate a potential framework to address these challenges specifically, while evaluating a human settlement detection task. A novel binary search tree sampling scheme is fused with a kernel based hashing procedure that maps image patches into hash-buckets using binary codes generated from image content. The framework exploits inherent redundancy within billions of image patches to promote mostly high variance preserving samples for accelerating algorithmic training and increasing model generalization.

Lunga, Dalton↗

Combining Sparse Approximate Factorizations with Mixed-precision Iterative Refinement

The standard LU factorization-based solution process for linear systems can be enhanced in speed or accuracy by employing mixed-precision iterative refinement. Most recent work has focused on dense systems. We investigate the potential of mixed-precision iterative refinement to enhance methods for sparse systems based on approximate sparse factorizations. In doing so, we first develop a new error analysis for LU- and GMRES-based iterative refinement under a general model of LU factorization that accounts for the approximation methods typically used by modern sparse solvers, such as low-rank approximations or relaxed pivoting strategies. We then provide a detailed performance analysis of both the execution time and memory consumption of different algorithms, based on a selected set of iterative refinement variants and approximate sparse factorizations. Our performance study uses the multifrontal solver MUMPS, which can exploit block low-rank factorization and static pivoting. We evaluate the performance of the algorithms on large, sparse problems coming from a variety of real-life and industrial applications showing that mixed-precision iterative refinement combined with approximate sparse factorization can lead to considerable reductions of both the time and memory consumption.

97 MATHEMATICS AND COMPUTING↗

SU-F-BRA-13: Knowledge-Based Treatment Planning for Prostate LDR Brachytherapy Based On Principle Component Analysis

Purpose: To create a knowledge-based algorithm for prostate LDR brachytherapy treatment planning that standardizes plan quality using seed arrangements tailored to individual physician preferences while being fast enough for real-time planning. Methods: A dataset of 130 prior cases was compiled for a physician with an active prostate seed implant practice. Ten cases were randomly selected to test the algorithm. Contours from the 120 library cases were registered to a common reference frame. Contour variations were characterized on a point by point basis using principle component analysis (PCA). A test case was converted to PCA vectors using the same process and then compared with each library case using a Mahalanobis distance to evaluate similarity. Rank order PCA scores were used to select the best-matched library case. The seed arrangement was extracted from the best-matched case and used as a starting point for planning the test case. Computational time was recorded. Any subsequent modifications were recorded that required input from a treatment planner to achieve an acceptable plan. Results: The computational time required to register contours from a test case and evaluate PCA similarity across the library was approximately 10s. Five of the ten test cases did not require any seed additions, deletions, or moves to obtain an acceptable plan. The remaining five test cases required on average 4.2 seed modifications. The time to complete manual plan modifications was less than 30s in all cases. Conclusion: A knowledge-based treatment planning algorithm was developed for prostate LDR brachytherapy based on principle component analysis. Initial results suggest that this approach can be used to quickly create treatment plans that require few if any modifications by the treatment planner. In general, test case plans have seed arrangements which are very similar to prior cases, and thus are inherently tailored to physician preferences.

60 APPLIED LIFE SCIENCES↗

Application of Backward Differentiation Formulas to Neutron Kinetics Problems

The neutron kinetics equations belong to the class of stiff equations for numerical time integration schemes. In this work, the accuracy and speed of algorithms based on backward differentiation formulas (BDFs) are studied with regard to point and spatial neutron kinetics problems. Using a BDF algorithm with the automatic selection of time step and order, solutions to a number of model problems with both a positive and negative reactivity insertion are analyzed. Plots of numerical cost dependences on local and global errors are presented. The results indicate that algorithms based on BDFs are highly efficient and that their application to nuclear reactor simulation is justified.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Parallel processing of filtered queries in attributed semantic graphs

Execution of complex analytic queries on massive semantic graphs is a challenging problem in big-data analytics that requires high-performance parallel computing. In a semantic graph, vertices and edges carry attributes of various types and the analytic queries typically depend on the values of these attributes. Thus, the computation must view the graph through a filter that passes only those individual vertices and edges of interest. Previous investigations have developed Knowledge Discovery Toolbox (KDT), a sophisticated Python library for parallel graph computations. In KDT, the user can write custom graph algorithms by specifying operations between edges and vertices ( semiring operations ). The user can also customize existing graph algorithms by writing filters. While the high-level language for this customization enables domain scientists to productively express their graph analytics requirements, the customized queries perform poorly due to the overhead of having to call into the Python virtual machine for each vertex and edge. In this work, we use the Selective Embedded Just-In-Time Specialization (SEJITS) approach to automatically translate semiring operations and filters defined by programmers into a lower-level efficiency language, bypassing the upcall into Python. We evaluate our approach by comparing it with the high-performance Combinatorial BLAS engine and show that our approach combines the benefits of programming in a high-level language with executing in a low-level parallel environment. We increase the system's flexibility by developing techniques that provide users with the ability to define new vertex and edge types from Python. We also present a new Roofline model for graph traversals and show that we achieve performance that is significantly closer to the bounds suggested by the Roofline. Finally, to further understand the complex interaction with the underlying architecture, we present an analysis using performance counters that quantifies the improvement in hardware behavior in the context our SEJITS methodology. Comprehensively, we demonstrate the first known solution to the problem of obtaining high performance from a productivity language when applying graph algorithms selectively on semantic graphs with hundreds of millions of edges and scaling to thousands of processors for graphs.

97 MATHEMATICS AND COMPUTING↗

Frequency-guided windowed Fourier ridges technique for automatic demodulation of a single closed fringe pattern

A frequency-guided windowed Fourier ridges (FG-WFR) technique for phase extraction from a single closed fringe pattern is proposed. The algorithm begins with a selected seed point and is then guided by the local frequency of the computed pixels. For the seed point, the whole spectral range is examined with WFR to provide reliable local frequency and phase estimation. However, for the rest of the pixels, the WFR algorithm is performed in a more narrow spectral range centered at the determined frequency of its computed neighboring pixel, which not only enforces the continuity of the local frequency between adjacent pixels but also reduces computation consumption significantly. Besides, since phase unwrapping can be readily implemented after demodulation of each pixel, the algorithm directly offers a continuous phase map. The proposed FG-WFR algorithm is verified by successfully demodulating two computer simulated fringe patterns and one noisy experimental fringe pattern.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗