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 1,333 records · Page 74

Predicting novel substrates for enzymes with minimal experimental effort with active learning

Enzymatic substrate promiscuity is more ubiquitous than previously thought, with significant consequences for understanding metabolism and its application to biocatalysis. This realization has given rise to the need for efficient characterization of enzyme promiscuity. Enzyme promiscuity is currently characterized with a limited number of human-selected compounds that may not be representative of the enzyme's versatility. While testing large numbers of compounds may be impractical, computational approaches can exploit existing data to determine the most informative substrates to test next, thereby more thoroughly exploring an enzyme's versatility. Here, to demonstrate this, we used existing studies and tested compounds for four different enzymes, developed support vector machine (SVM) models using these datasets, and selected additional compounds for experiments using an active learning approach. SVMs trained on a chemically diverse set of compounds were discovered to achieve maximum accuracies of similar to 80% using similar to 33% fewer compounds than datasets based on all compounds tested in existing studies. Active learning-selected compounds for testing resolved apparent conflicts in the existing training data, while adding diversity to the dataset. Finally, the application of these algorithms to wide arrays of metabolic enzymes would result in a library of SVMs that can predict high-probability promiscuous enzymatic reactions and could prove a valuable resource for the design of novel metabolic pathways.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Advances in molecular quantum chemistry contained in the Q-Chem 4 program package

A summary of the technical advances that are incorporated in the fourth major release of the Q-Chem quantum chemistry program is provided, covering approximately the last seven years. These include developments in density functional theory methods and algorithms, nuclear magnetic resonance (NMR) property evaluation, coupled cluster and perturbation theories, methods for electronically excited and open-shell species, tools for treating extended environments, algorithms for walking on potential surfaces, analysis tools, energy and electron transfer modelling, parallel computing capabilities, and graphical user interfaces. In addition, a selection of example case studies that illustrate these capabilities is given. These include extensive benchmarks of the comparative accuracy of modern density functionals for bonded and non-bonded interactions, tests of attenuated second order Møller-Plesset (MP2) methods for intermolecular interactions, a variety of parallel performance benchmarks, and tests of the accuracy of implicit solvation models. Finally, some specific chemical examples include calculations on the strongly correlated Cr 2 dimer, exploring zeolite-catalysed ethane dehydrogenation, energy decomposition analysis of a charged ter-molecular complex arising from glycerol photoionisation, and natural transition orbitals for a Frenkel exciton state in a nine-unit model of a self-assembling nanotube.

computational modelling↗

Scanning electron microscopy image representativeness: morphological data on nanoparticles

Summary A sample of a nanomaterial contains a distribution of nanoparticles of various shapes and/or sizes. A scanning electron microscopy image of such a sample often captures only a fragment of the morphological variety present in the sample. In order to quantitatively analyse the sample using scanning electron microscope digital images, and, in particular, to derive numerical representations of the sample morphology, image content has to be assessed. In this work, we present a framework for extracting morphological information contained in scanning electron microscopy images using computer vision algorithms, and for converting them into numerical particle descriptors. We explore the concept of image representativeness and provide a set of protocols for selecting optimal scanning electron microscopy images as well as determining the smallest representative image set for each of the morphological features. We demonstrate the practical aspects of our methodology by investigating tricalcium phosphate, Ca 3 (PO 4 ) 2 , and calcium hydroxyphosphate, Ca 5 (PO 4 ) 3 (OH), both naturally occurring minerals with a wide range of biomedical applications.

ODZIOMEK, KATARZYNA↗

MO-DE-207B-08: Radiomic CT Features Complement Semantic Annotations to Predict EGFR Mutations in Lung Adenocarcinomas

Purpose: To compare the complementary value of quantitative radiomic features to that of radiologist-annotated semantic features in predicting EGFR mutations in lung adenocarcinomas. Methods: Pre-operative CT images of 258 lung adenocarcinoma patients were available. Tumors were segmented using the sing-click ensemble segmentation algorithm. A set of radiomic features was extracted using 3D-Slicer. Test-retest reproducibility and unsupervised dimensionality reduction were applied to select a subset of reproducible and independent radiomic features. Twenty semantic annotations were scored by an expert radiologist, describing the tumor, surrounding tissue and associated findings. Minimum-redundancy-maximum-relevance (MRMR) was used to identify the most informative radiomic and semantic features in 172 patients (training-set, temporal split). Radiomic, semantic and combined radiomic-semantic logistic regression models to predict EGFR mutations were evaluated in and independent validation dataset of 86 patients using the area under the receiver operating curve (AUC). Results: EGFR mutations were found in 77/172 (45%) and 39/86 (45%) of the training and validation sets, respectively. Univariate AUCs showed a similar range for both feature types: radiomics median AUC = 0.57 (range: 0.50 – 0.62); semantic median AUC = 0.53 (range: 0.50 – 0.64, Wilcoxon p = 0.55). After MRMR feature selection, the best-performing radiomic, semantic, and radiomic-semantic logistic regression models, for EGFR mutations, showed a validation AUC of 0.56 (p = 0.29), 0.63 (p = 0.063) and 0.67 (p = 0.004), respectively. Conclusion: Quantitative volumetric and textural Radiomic features complement the qualitative and semi-quantitative radiologist annotations. The prognostic value of informative qualitative semantic features such as cavitation and lobulation is increased with the addition of quantitative textural features from the tumor region.

60 APPLIED LIFE SCIENCES↗

Search for direct top squark pair production in final states with one isolated lepton, jets, and missing transverse momentum in s = 7 TeV p p collisions using 4.7 f b − 1 of ATLAS data

CERN-LHC. A study of data from proton-proton collisions at a centre-of-mass energy of 7 TeV having final states consisting of one isolated electron or muon, jets and missing trasnverse momentum using a data sample of total integrated luminosity of 4.7 fb-1. Each event must have >=1 reconstructed primary vertex associated with five or more tracks with PT > 0.4 GeV. Events must have exactly one muon (or electron) with |eta|<2.4 (or 2.47) and pT > 20 (or 25) GeV and isolated from other particles. The scalar sum of the pT of tracks >1 GeV within a cone of size deltaR=0.2 around the lepton candidate is required to be < 10% of the electron pT, and <1.8 Gev for the muon. jets are reconstructed using the ant-kT clustering algorithm with radius parameter=0.4. Events with four or more jets with |eta|<2.5 and pT> 80,60,40,and 25 GeV are selected with at least one jet identified as a b-jet.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

CEED-MS8:Initial Integration of CEED Software in ECP Applications

Key components of CEED software development involve fast finite element operator storage and evaluation, architecture optimizations, performant algorithms for all orders, global kernels for finite element operators, efficient use of the memory sub-system, optimal data locality and motion, enhanced scalability and parallelism, and fast tensor contractions. As ultimate goals of the project, CEED is exploring and identifying the best algorithms for the full range of discretizations and applying algorithmic and software development to support ECP applications’ needs. In this milestone we report on the initial integration of CEED software in the selected first-wave ECP/CEED apps and the continued exploration of other ECP applications as second-wave (milestone CEED-MS23) and third-wave (milestone CEED-MS35) ECP/CEED application candidates. An extensive evaluation of various approaches for GPU acceleration of the CEED computational motives is also a major focus of the report, both in the settings of CEED’s Nekbone and Laghos miniapps, as well as versions of CEED’s bake-off problems (introduced in milestone CEED-MS6). The first-wave ECP applications targeted for coupling with CEED software include the ExaSMR and MARBL applications. We also report on near-term collaboration with the seed-funded Urban Systems project, as well as engagements with ACME, ExaAM, GEOS and the Application Assessment and Proxy App projects. The ExaSMR project involves coupled thermal-hydraulics/neutronics, with the latter based on Monte Carlo methods and the former based on high-order discretizations of the Navier-Stokes and energy equations for accurate simulation of turbulent transport. The project requires fast and efficient turbulence simulation capabilities that will scale to exascale platforms. Work within CEED will ensure that all aspects of the Nek5000 workflow scale to the target problem space, which involves > 10 10 degrees of freedom in typical production runs. CEED’s goal is to provide implementations that perform with maximum attainable efficiency and deliver low turn-around times. A deliverable for this milestone is beta-release of the GPU-based variant of Nek5000. MARBL is targeting high-energy-density physics problems, including single fluid multi-material hydrodynamics and radiation/magnetic diffusion simulation, with applications in inertial confined fusion, pulsed power experiments, and equation of state/material strength analysis. The code uses high-order methods based arbitrary Lagrangian-Eulerian, direct Eulerian, and unstructured adaptive mesh refinement. A deliverable for this milestone is collaboration with MARBL for improved performance based on partially-assembled operators. The Urban exascale project is planning to couple real-time data with high performance simulations to study particulate transport in cities. Nek5000 will be used for large eddy and RANS (Reynolds-averaged Navier-Stokes) simulations of turbulent transport in these domains. Areas where the CEED team is aiding the Urban Canyon group include creating high-order meshes for urban geometries and developing boundary conditions appropriate for simulations close to real-life urban conditions. The software artifacts delivered as part of this milestone include GPU-enabled versions of the Nekbone and Laghos miniapps, which are provided through the CEED website, http://ceed.exascaleproject.org and the CEED GitHub organization, http://github.com/ceed. In addition to details and results from the above R&D efforts, in this document we are also reporting on other project-wide activities performed in Q4 of FY17, including: the CEED first annual meeting and participation of CEED researchers in a variety of outreach activities.

97 MATHEMATICS AND COMPUTING↗

Introduction to ALERT: A User-Interface Tool for Resilient Optimization

As the cyber-physical systems grow in complexity, there is a need for proactive resilience strategies that involve online, adaptive control actions to best prepare for any impending adversarial events. In this technical effort, supported by RD2C LDRD initiative, the project team designed and demonstrated online strategies – referred to as ALERT controls – for proactive and adaptive tuning of existing optimal controls in a microgrid, with quantifiably assured margins of resilience to various cyber-physical adversarial events. This ALERT functionality is made available to the end-users, e.g., the system operators, via an interactive user-interface. The end-users will not only be able to use the interface to visualize the system’s operation under various cyber-physical adversarial scenarios, but also evaluate the amount of tolerance the system has against selected adversarial perturbations of interest (e.g., malfunctioning sensors, suspected attacked measurements) via the adversarial plots. In this technical report, we briefly outline the algorithmic modules of the developed ALERT control technology, and introduce the user-interface tool that allows end-users (e.g., microgrid operators) to enter their system description, specify various operational and resilience requirements, and evaluate the impact of the control decisions via illustrative plots.

97 MATHEMATICS AND COMPUTING↗

Laser Based Ultrasound for Verification of Circuit Card Assemblies

This IR is a follow-up conference Paper to the IR approved abstract; PNNL-SA-194096. --- Verification that equipment is authentic and not changed; even at the circuit card assembly (CCA) level will likely be an important component of future arms control treaties. This effort was an initial evaluation of the potential for laser-based ultrasound (LBU) as an inspection tool for Unique Identification (UID) of Circuit Card Assembly (CCA) boards and CCA components. The LBU system used a laser pulse for ultrasound generation and an Optical Microphone for ultrasound detection to image subsurface structures without physical contact and in a dry state. This paper described the selected CCA surrogate, CCA components that were examined, the LBU system, LBU images of the CCA components, and the initial development of UID algorithms. Receiver operating characteristics (ROC) curves indicated good performance for two algorithms as a potential means for UID of CCA boards and CCA components.

Laser Ultrasound, Equipment Verification, Arms Con↗

Group Projected subspace pursuit for IDENTification of variable coefficient differential equations (GP-IDENT)

We propose an effective and robust algorithm for identifying partial differential equations (PDEs) with space-time varying coefficients from the noisy observation of a single solution trajectory. Identifying unknown differential equations from noisy data is a difficult task, and it is even more challenging with space and time varying coefficients in the PDE. The proposed algorithm, GP-IDENT, has three ingredients: (i) we use B-spline bases to express the unknown space and time varying coefficients, (ii) we propose Group Projected Subspace Pursuit (GPSP) to find a sequence of candidate PDEs with various levels of complexity, and (iii) we propose a new criterion for model selection using the Reduction in Residual (RR) to choose an optimal one among a pool of candidates. The new GPSP considers group projected subspaces which is more robust than existing methods in distinguishing correlated group features. We test GP-IDENT on a variety of PDEs and PDE systems, and compare it with the state-of-the-art parametric PDE identification algorithms under different settings to illustrate its outstanding performance. Furthermore, our experiments show that GP-IDENT is effective in identifying the correct terms from a large dictionary, and our model selection scheme is robust to noise.

Data-driven method↗

Traffic Control via Connected and Automated Vehicles (CAVs): An Open-Road Field Experiment with 100 CAVs

The CIRCLES project aims to reduce instabilities in traffic flow, which are naturally occurring phenomena due to human driving behavior. Also called “phantom jams” or “stop-and-go waves,” these instabilities are a significant source of wasted energy. Toward this goal, the CIRCLES project designed a control system, referred to as the MegaController by the CIRCLES team, that could be deployed in real traffic. Our field experiment, the MegaVanderTest (MVT), leveraged a heterogeneous fleet of 100 longitudinally controlled vehicles as Lagrangian traffic actuators, each of which ran a controller with the architecture described in this article. The MegaController is a hierarchical control architecture that consists of two main layers. The upper layer is called the Speed Planner and is a centralized optimal control algorithm. It assigns speed targets to the vehicles, conveyed through the LTE cellular network. The lower layer is a control layer, running on each vehicle. It performs local actuation by overriding the stock adaptive cruise controller, using the stock onboard sensors. The Speed Planner ingests live data feeds provided by third parties as well as data from our own control vehicles and uses both to perform the speed assignment. The architecture of the Speed Planner allows for the modular use of standard control techniques, such as optimal control, model predictive control (MPC), kernel methods, and others. The architecture of the local controller allows for the flexible implementation of local controllers. Corresponding techniques include deep reinforcement learning (RL), MPC, and explicit controllers. Depending on the vehicle architecture, all onboard sensing data can be accessed by the local controllers or only some. Likewise, control inputs vary across different automakers, with inputs ranging from torque or acceleration requests for some cars to electronic selection of adaptive cruise control (ACC) setpoints in others. The proposed architecture technically allows for the combination of all possible settings proposed previously, that is {Speed Planner algorithms} × {local Vehicle Controller algorithms} × {full or partial sensing} × {torque or speed control}. As a result, most configurations were tested throughout the ramp up to the MegaVandertest (MVT).

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Genome-wide functional screens enable the prediction of high activity CRISPR-Cas9 and -Cas12a guides in Yarrowia lipolytica

Abstract Genome-wide functional genetic screens have been successful in discovering genotype-phenotype relationships and in engineering new phenotypes. While broadly applied in mammalian cell lines and in E. coli , use in non-conventional microorganisms has been limited, in part, due to the inability to accurately design high activity CRISPR guides in such species. Here, we develop an experimental-computational approach to sgRNA design that is specific to an organism of choice, in this case the oleaginous yeast Yarrowia lipolytica . A negative selection screen in the absence of non-homologous end-joining, the dominant DNA repair mechanism, was used to generate single guide RNA (sgRNA) activity profiles for both SpCas9 and LbCas12a. This genome-wide data served as input to a deep learning algorithm, DeepGuide, that is able to accurately predict guide activity. DeepGuide uses unsupervised learning to obtain a compressed representation of the genome, followed by supervised learning to map sgRNA sequence, genomic context, and epigenetic features with guide activity. Experimental validation, both genome-wide and with a subset of selected genes, confirms DeepGuide’s ability to accurately predict high activity sgRNAs. DeepGuide provides an organism specific predictor of CRISPR guide activity that with retraining could be applied to other fungal species, prokaryotes, and other non-conventional organisms.

59 BASIC BIOLOGICAL SCIENCES↗

An adaptive stochastic sequential quadratic programming with differentiable exact augmented lagrangians

In this study, we consider solving nonlinear optimization problems with a stochastic objective and deterministic equality constraints. We assume for the objective that its evaluation, gradient, and Hessian are inaccessible, while one can compute their stochastic estimates by, for example, subsampling. We propose a stochastic algorithm based on sequential quadratic programming (SQP) that uses a differentiable exact augmented Lagrangian as the merit function. To motivate our algorithm design, we first revisit and simplify an old SQP method Lucidi developed for solving deterministic problems, which serves as the skeleton of our stochastic algorithm. Based on the simplified deterministic algorithm, we then propose a non-adaptive SQP for dealing with stochastic objective, where the gradient and Hessian are replaced by stochastic estimates but the stepsizes are deterministic and prespecified. Finally, we incorporate a recent stochastic line search procedure Paquette and Scheinberg into the non-adaptive stochastic SQP to adaptively select the random stepsizes, which leads to an adaptive stochastic SQP. The global "almost sure" convergence for both non-adaptive and adaptive SQP methods is established. Numerical experiments on nonlinear problems in CUTEst test set demonstrate the superiority of the adaptive algorithm.

97 MATHEMATICS AND COMPUTING↗

Interactive prostate segmentation using atlas-guided semi-supervised learning and adaptive feature selection

Purpose: Accurate prostate segmentation is necessary for maximizing the effectiveness of radiation therapy of prostate cancer. However, manual segmentation from 3D CT images is very time-consuming and often causes large intra- and interobserver variations across clinicians. Many segmentation methods have been proposed to automate this labor-intensive process, but tedious manual editing is still required due to the limited performance. In this paper, the authors propose a new interactive segmentation method that can (1) flexibly generate the editing result with a few scribbles or dots provided by a clinician, (2) fast deliver intermediate results to the clinician, and (3) sequentially correct the segmentations from any type of automatic or interactive segmentation methods. Methods: The authors formulate the editing problem as a semisupervised learning problem which can utilize a priori knowledge of training data and also the valuable information from user interactions. Specifically, from a region of interest near the given user interactions, the appropriate training labels, which are well matched with the user interactions, can be locally searched from a training set. With voting from the selected training labels, both confident prostate and background voxels, as well as unconfident voxels can be estimated. To reflect informative relationship between voxels, location-adaptive features are selected from the confident voxels by using regression forest and Fisher separation criterion. Then, the manifold configuration computed in the derived feature space is enforced into the semisupervised learning algorithm. The labels of unconfident voxels are then predicted by regularizing semisupervised learning algorithm. Results: The proposed interactive segmentation method was applied to correct automatic segmentation results of 30 challenging CT images. The correction was conducted three times with different user interactions performed at different time periods, in order to evaluate both the efficiency and the robustness. The automatic segmentation results with the original average Dice similarity coefficient of 0.78 were improved to 0.865–0.872 after conducting 55–59 interactions by using the proposed method, where each editing procedure took less than 3 s. In addition, the proposed method obtained the most consistent editing results with respect to different user interactions, compared to other methods. Conclusions: The proposed method obtains robust editing results with few interactions for various wrong segmentation cases, by selecting the location-adaptive features and further imposing the manifold regularization. The authors expect the proposed method to largely reduce the laborious burdens of manual editing, as well as both the intra- and interobserver variability across clinicians.

07 ISOTOPE AND RADIATION SOURCES↗

Fitting coupled potential energy surfaces for large systems: Method and construction of a 3-state representation for phenol photodissociation in the full 33 internal degrees of freedom using multireference configuration interaction determined data

A recently reported algorithm for representing adiabatic states coupled by conical intersections using a quasi-diabatic state Hamiltonian in four and five atom systems is extended to treat nonadiabatic processes in considerably larger molecules. The method treats all internal degrees of freedom and uses electronic structure data from ab initio multireference configuration interaction wave functions with nuclear configuration selection based on quasi-classical surface hopping trajectories. The method is shown here to be able to treat ∼30 internal degrees of freedom including dissociative and large amplitude internal motion. Two procedures are introduced which are essential to the algorithm, a null space projector which removes basis functions from the fitting process until they are needed and a partial diagonalization technique which allows for automated, but accurate, treatment of the vicinity of extended seams of conical intersections of two or more states. These procedures are described in detail. The method is illustrated using the photodissociaton of phenol, C{sub 6}H{sub 5}OH(X{sup ~1}A{sup ′}) + hv → C{sub 6}H{sub 5}OH(A{sup ~1}A{sup ′}, B{sup ~1}A{sup ′′}) → C{sub 6}H{sub 5}O(X{sup ~2}B{sub 1}, A{sup ~2}B{sub 2}) + H as a test case. Ab initio electronic structure data for the 1,2,3{sup 1}A states of phenol, which are coupled by conical intersections, are obtained from multireference first order configuration interaction wave functions. The design of bases to simultaneously treat large amplitude motion and dissociation is described, as is the ability of the fitting procedure to smooth the irregularities in the electronic energies attributable to the orbital changes that are inherent to nonadiabatic processes.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Noise suppression in reconstruction of low-Z target megavoltage cone-beam CT images

Purpose: To improve the image contrast-to-noise (CNR) ratio for low-Z target megavoltage cone-beam CT (MV CBCT) using a statistical projection noise suppression algorithm based on the penalized weighted least-squares (PWLS) criterion. Methods: Projection images of a contrast phantom, a CatPhan{sup Registered-Sign} 600 phantom and a head phantom were acquired by a Varian 2100EX LINAC with a low-Z (Al) target and low energy x-ray beam (2.5 MeV) at a low-dose level and at a high-dose level. The projections were then processed by minimizing the PWLS objective function. The weighted least square (WLS) term models the noise of measured projection and the penalty term enforces the smoothing constraints of the projection image. The variance of projection data was chosen as the weight for the PWLS objective function and it determined the contribution of each measurement. An anisotropic quadratic form penalty that incorporates the gradient information of projection image was used to preserve edges during noise reduction. Low-Z target MV CBCT images were reconstructed by the FDK algorithm after each projection was processed by the PWLS smoothing. Results: Noise in low-Z target MV CBCT images were greatly suppressed after the PWLS projection smoothing, without noticeable sacrifice of the spatial resolution. Depending on the choice of smoothing parameter, the CNR of selected regions of interest in the PWLS processed low-dose low-Z target MV CBCT image can be higher than the corresponding high-dose image.Conclusion: The CNR of low-Z target MV CBCT images was substantially improved by using PWLS projection smoothing. The PWLS projection smoothing algorithm allows the reconstruction of high contrast low-Z target MV CBCT image with a total dose of as low as 2.3 cGy.

60 APPLIED LIFE SCIENCES↗

Improving the chances of successful protein structure determination with a random forest classifier

Using an extended set of protein features calculated separately for protein surface and interior, a new version of XtalPred based on a random forest classifier achieves a significant improvement in predicting the success of structure determination from the primary amino-acid sequence. Obtaining diffraction quality crystals remains one of the major bottlenecks in structural biology. The ability to predict the chances of crystallization from the amino-acid sequence of the protein can, at least partly, address this problem by allowing a crystallographer to select homologs that are more likely to succeed and/or to modify the sequence of the target to avoid features that are detrimental to successful crystallization. In 2007, the now widely used XtalPred algorithm [Slabinski et al. (2007 ▶), Protein Sci.16, 2472–2482] was developed. XtalPred classifies proteins into five ‘crystallization classes’ based on a simple statistical analysis of the physicochemical features of a protein. Here, towards the same goal, advanced machine-learning methods are applied and, in addition, the predictive potential of additional protein features such as predicted surface ruggedness, hydrophobicity, side-chain entropy of surface residues and amino-acid composition of the predicted protein surface are tested. The new XtalPred-RF (random forest) achieves significant improvement of the prediction of crystallization success over the original XtalPred. To illustrate this, XtalPred-RF was tested by revisiting target selection from 271 Pfam families targeted by the Joint Center for Structural Genomics (JCSG) in PSI-2, and it was estimated that the number of targets entered into the protein-production and crystallization pipeline could have been reduced by 30% without lowering the number of families for which the first structures were solved. The prediction improvement depends on the subset of targets used as a testing set and reaches 100% (i.e. twofold) for the top class of predicted targets.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Co-design of Advanced Architectures for Graph Analytics using Machine Learning

A graph is an excellent way of representing relationships among entities. We can use graph analytics to synthesize and analyze such relational data, and extract relevant features that are useful for various tasks such as machine learning. Considering the crucial role of graph analytics in various domains, it is important and timely to investigate the right hardware configurations that can achieve optimal performance for graph workloads on future high-performance computing systems. Design space exploration studies facilitate the selection of appropriate configurations (e.g. memory) to achieve a desired system performance. Recently, the approach of accelerating graph analytics using persistent non-volatile memory has gained a lot of attention. Traditional system simulators such as Gem5 and NVMain can be used to explore the design space of these advanced memory architectures for graph workloads. However, these simulators are slow in execution thus limiting the efficiency of design space exploration studies. To overcome this challenge, we proposed a machine learning based approach to co-design advanced memory architectures for graph workloads. We tested our approach with DRAM, non-volatile memory, and hybrid memory (DRAM+NVM) using a breadth first search benchmark algorithm. Our results showed the applicability of the proposed machine learning based approach to the co-design of the advanced memory architectures. In this paper, we provide recommendations on selecting advanced memory architectures to achieve desired performance for graph workloads. We also discuss the performances of different machine learning models that were considered in this study.

Kurte, Kuldeep↗

WinnowML: Stable feature selection for maximizing prediction accuracy of time-based system modeling

Online deep learning (ODL) has become an important methodology for modeling time-based performance of computer systems. An open problem is the intelligent selection of features from raw workload traces of computer systems. The best methods are overly sensitive to noisy data, causing frequent feature changes and re-training. Using all available features inflates training time and introduces model artifacts if some features should have been dropped. We present WinnowML, a method for automatically determining the most relevant feature subset for a predictive time-series model. WinnowML combines existing feature ranking algorithms and a history of each feature's ranking to iteratively rank a feature set to lower prediction error and maximize long term relevance. From this ranked feature set, the most relevant and stable subset is selected to train a model. Experimentally, we show how WinnowML can lower a model's mean absolute relative error up to 42% on average compared to the closest performing approach. Additionally, we lower the fluctuation in feature ranking and selection up to 65%. We also demonstrate how to combine WinnowML and a model search tool to provide improvements in performance of up to 14.5% when compared to using all the feature available.

Bel, Oceane MS↗