Search NASA⌕ Search

SEARCH · Search NASA

Results for “Structure preserving 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.

At least 199 records · Page 11

Anatomical pulmonary magnetic resonance imaging segmentation for regional structure-function measurements of asthma

Purpose: Pulmonary magnetic-resonance-imaging (MRI) and x-ray computed-tomography have provided strong evidence of spatially and temporally persistent lung structure-function abnormalities in asthmatics. This has generated a shift in their understanding of lung disease and supports the use of imaging biomarkers as intermediate endpoints of asthma severity and control. In particular, pulmonary {sup 1}H MRI can be used to provide quantitative lung structure-function measurements longitudinally and in response to treatment. However, to translate such biomarkers of asthma, robust methods are required to segment the lung from pulmonary {sup 1}H MRI. Therefore, their objective was to develop a pulmonary {sup 1}H MRI segmentation algorithm to provide regional measurements with the precision and speed required to support clinical studies. Methods: The authors developed a method to segment the left and right lung from {sup 1}H MRI acquired in 20 asthmatics including five well-controlled and 15 severe poorly controlled participants who provided written informed consent to a study protocol approved by Health Canada. Same-day spirometry and plethysmography measurements of lung function and volume were acquired as well as {sup 1}H MRI using a whole-body radiofrequency coil and fast spoiled gradient-recalled echo sequence at a fixed lung volume (functional residual capacity + 1 l). We incorporated the left-to-right lung volume proportion prior based on the Potts model and derived a volume-proportion preserved Potts model, which was approximated through convex relaxation and further represented by a dual volume-proportion preserved max-flow model. The max-flow model led to a linear problem with convex and linear equality constraints that implicitly encoded the proportion prior. To implement the algorithm, {sup 1}H MRI was resampled into ∼3 × 3 × 3 mm{sup 3} isotropic voxel space. Two observers placed seeds on each lung and on the background of 20 pulmonary {sup 1}H MR images in a randomized dataset, on five occasions, five consecutive days in a row. Segmentation accuracy was evaluated using the Dice-similarity-coefficient (DSC) of the segmented thoracic cavity with comparison to five-rounds of manual segmentation by an expert observer. The authors also evaluated the root-mean-squared-error (RMSE) of the Euclidean distance between lung surfaces, the absolute, and percent volume error. Reproducibility was measured using the coefficient of variation (CoV) and intraclass correlation coefficient (ICC) for two observers who repeated segmentation measurements five-times. Results: For five well-controlled asthmatics, forced expiratory volume in 1 s (FEV{sub 1})/forced vital capacity (FVC) was 83% ± 7% and FEV{sub 1} was 86 ± 9%{sub pred}. For 15 severe, poorly controlled asthmatics, FEV{sub 1}/FV C = 66% ± 17% and FEV{sub 1} = 72 ± 27%{sub pred}. The DSC for algorithm and manual segmentation was 91% ± 3%, 92% ± 2% and 91% ± 2% for the left, right, and whole lung, respectively. RMSE was 4.0 ± 1.0 mm for each of the left, right, and whole lung. The absolute (percent) volume errors were 0.1 l (∼6%) for each of right and left lung and ∼0.2 l (∼6%) for whole lung. Intra- and inter-CoV (ICC) were <0.5% (>0.91%) for DSC and <4.5% (>0.93%) for RMSE. While segmentation required 10 s including ∼6 s for user interaction, the smallest detectable difference was 0.24 l for algorithm measurements which was similar to manual measurements. Conclusions: This lung segmentation approach provided the necessary and sufficient precision and accuracy required for research and clinical studies.

60 APPLIED LIFE SCIENCES↗

Nonradiative relaxation dynamics in the [Au 25- n Ag n (SH) 18 ] -1 ( n = 1, 12, 25) thiolate-protected nanoclusters

Evaluation of the electron-nuclear dynamics and relaxation mechanisms of gold and silver nanoclusters and their alloys is important for future photocatalytic, light harvesting, and photoluminescence applications of these systems. In this work, the effect of silver doping on the nonradiative excited state relaxation dynamics of the atomically precise thiolate-protected gold nanocluster [Au 25- n Ag n (SH) 18 ] –1 ( n = 1, 12, 25) is studied theoretically. Time-dependent density functional theory is used to study excited states lying in the energy range 0.0–2.5 eV. The fewest switches surface hopping method with decoherence correction was used to investigate the dynamics of these states. The HOMO–LUMO gap increases significantly upon doping of 12 silver atoms but decreases for the pure silver nanocluster. Doped clusters show a different response for ground state population increase lifetimes and excited state population decay times in comparison to the undoped system. The ground state recovery times of the S 1 –S 6 states in the first excited peak were found to be longer for [Au 13 Ag 12 (SH) 18 ] –1 than the corresponding recovery times of other studied nanoclusters, suggesting that this partially doped nanocluster is best for preserving electrons in an excited state. The decay time constants were in the range of 2.0–20 ps for the six lowest energy excited states. Among the higher excited states, S 7 has the slowest decay time constant although it occurs more quickly than S 1 decay. Overall, these clusters follow common decay time constant trends and relaxation mechanisms due to the similarities in their electronic structures.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Structural analysis of cortical porosity applied to HR-pQCT data

Purpose: The investigation of cortical porosity is an important aspect of understanding biological, pathoetiological, and biomechanical processes occurring within the skeleton. With the emergence of HR-pQCT as a noninvasive tool suitable for clinical use, cortical porosity at appendicular sites can be directly visualizedin vivo. The aim of this study was to introduce a novel topological analysis of the cortical pore network for HR-pQCT data and determine the influence of resolution on measures of cortical pore network microstructure and topology. Methods: Cadaveric radii were scanned using HR-pQCT at two different voxel sizes (41 and 82μm) and also using μCT at a voxel size of 18 μm. HR-pQCT and μCT image sets were spatially coregistered. Segmentation and quantification of cortical porosity (Ct.Po) and mean pore diameter (Ct.Po.Dm) were achieved using an established extended cortical analysis technique. Topological classification of individual pores was performed using topology-preserving skeletonization and multicolor dilation algorithms. Based on the pore skeleton topological classification, the following parameters were quantified: total number of planar surface-skeleton canals (N.Slabs), tubular curve-skeleton canals (N.Tubes), and junction elements (N.Junctions), mean slab volume (Slab.Vol), mean tube volume (Tube.Vol), mean slab orientation (Slab.θ), mean tube orientation (Tube.θ), N.Slabs/N.Tubes, and integral (total) slab volume/integral tube volume (iSlab.Vol/iTube.Vol). An in vivo reproducibility study was also conducted to assess short-term precision of the topology parameters. Precision error was characterized using root mean square coefficient of variation (RMSCV%). Results: Correlations toμCT values for Ct.Po were significant for both the 41 and 82 μm HR-pQCT data (41: r{sup 2} = 0.82, p < 0.001, 82: r{sup 2} = 0.75, p < 0.001). For Ct.Po.Dm, only the 41 μm data were significantly predictive of μCT values (r{sup 2} = 0.72, p < 0.01) Data at both HR-pQCT voxel sizes were strongly predictive of the μCT values for N.Slabs (41: r{sup 2} = 0.93, p < 0.001; 82: r{sup 2} = 0.84, p < 0.001), N.Tubes (41: r{sup 2} = 0.94, p < 0.001; 82: r{sup 2} = 0.84, p < 0.001), and N.Junctions (41: r{sup 2} = 0.93, p < 0.001; 82: r{sup 2} = 0.78, p < 0.001), though proportional bias was evident in these correlations. Weak correlations were seen for iSlab.Vol/iTube.Vol at both voxel sizes (41: r{sup 2} = 0.52, p < 0.01; 82: r{sup 2} = 0.39, p < 0.05). Slab.Vol was significantly correlated to μCT data at 41 μm (r{sup 2} = 0.60, p < 0.01) but not at 82 μm, while Tube.Vol was significantly correlated at both voxel sizes (41: r{sup 2} = 0.79, p < 0.001; 82: r{sup 2} = 0.68, p < 0.01). In vivo precision error for these parameters ranged from 2.31 to 9.68 RMSCV%. Conclusions: Strong correlations betweenμCT- and HR-pQCT-derived measurements were found, particularly in HR-pQCT images obtained at 41 μm. These data are in agreement with our previous study investigating the effect of voxel size on standard HR-pQCT metrics of trabecular and cortical microstructure, and extend our previous findings to include topological descriptors of the cortical pore network.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Calculating the Symmetry Number of Flexible Sphere Clusters

We present a theoretical and computational framework to compute the symmetry number of a flexible sphere cluster in R 3 , using a definition of symmetry that arises naturally when calculating the equilibrium probability of a cluster of spheres in the sticky-sphere limit. We define the sticky symmetry group of the cluster as the set of permutations and inversions of the spheres which preserve adjacency and can be realized by continuous deformations of the cluster that do not change the set of contacts or cause particles to overlap. The symmetry number is the size of the sticky symmetry group. We introduce a numerical algorithm to compute the sticky symmetry group and symmetry number, and show it works well on several test cases. Furthermore, we show that once the sticky symmetry group has been calculated for indistinguishable spheres, the symmetry group for partially distinguishable spheres (those with nonidentical interactions) can be efficiently obtained without repeating the laborious parts of the computations. Finally, we use our algorithm to calculate the partition functions of every possible connected cluster of six identical sticky spheres, generating data that may be used to design interactions between spheres so they self-assemble into a desired structure.

97 MATHEMATICS AND COMPUTING↗

Efficient graph representation framework for chemical molecule similarity tasks

Graph data has emerged in numerous scientific domains and machine learning techniques have been widely used for analysis and learning of diverse data for prediction and decision. Machine learning techniques can readily address complex problems by leveraging their structural information. But graphs cannot be directly used for existing machine learning algorithms unless encoded as vectors. The problem of efficient representation of graphs is a substantial challenge in graph machine learning. In this paper, we propose a novel two-stage framework for the representation of chemical molecule graphs based on the strengths of Graph Isomorphism Networks (GINs) and Siamese autoencoders. In the first stage, the GIN model is constructed and trained using the structural information of chemical molecule graphs. Node attributes, edge attributes, and edge indices are used as input data, while graph attributes are used as labels. The GIN model effectively captures the structural characteristics of graphs and can accurately predict graph attributes, i.e., molecular properties. It also generates Graph Embeddings, represented as vectors that encode the structural information of graphs. In the second stage, Graph Embedding vectors are further optimized for downstream similarity tasks while preserving the graph structural information. The Siamese autoencoder is constructed and trained, which reduces the dimensionality of the Graph Embedding vectors, while maximizing the preservation of structural information in the original high-dimensional vectors. The resulting low-dimensional Graph Embeddings can be effectively utilized for tasks such as approximate nearest neighbor search. The experimental results demonstrate the effectiveness of our proposed framework in accurately predicting graph similarity.

Ma, Jiaji↗

Survey and analysis of multiresolution methods for turbulence data

This paper compares the effectiveness of various multi-resolution geometric representation methods, such as B-spline, Daubechies, Coiflet and Dual-tree wavelets, curvelets and surfacelets, to capture the structure of fully developed turbulence using a truncated set of coefficients. The turbulence dataset is obtained from a Direct Numerical Simulation of buoyancy driven turbulence on a 512 3 mesh size, with an Atwood number, A = 0.05, and turbulent Reynolds number, Re t = 1800, and the methods are tested against quantities pertaining to both velocities and active scalar (density) fields and their derivatives, spectra, and the properties of constant density surfaces. The comparisons between the algorithms are given in terms of performance, accuracy, and compression properties. The results should provide useful information for multi-resolution analysis of turbulence, coherent feature extraction, compression for large datasets handling, as well as simulations algorithms based on multi-resolution methods. In conclusion, the final section provides recommendations for best decomposition algorithms based on several metrics related to computational efficiency and preservation of turbulence properties using a reduced set of coefficients.

97 MATHEMATICS AND COMPUTING↗

Data-driven modeling and control of an X-ray bimorph adaptive mirror

Adaptive X-ray mirrors are being adopted on high-coherent-flux synchrotron and X-ray free-electron laser beamlines where dynamic phase control and aberration compensation are necessary to preserve wavefront quality from source to sample, yet challenging to achieve. Additional difficulties arise from the inability to continuously probe the wavefront in this context, which demands methods of control that require little to no feedback. In this work, a data-driven approach to the control of adaptive X-ray optics with piezo-bimorph actuators is demonstrated. This approach approximates the non-linear system dynamics with a discrete-time model using random mirror shapes and interferometric measurements as training data. For mirrors of this type, prior states and voltage inputs affect the shape-change trajectory, and therefore must be included in the model. Without the need for assumed physical models of the mirror's behavior, the generality of the neural network structure accommodates drift, creep and hysteresis, and enables a control algorithm that achieves shape control and stability below 2 nm RMS. Using a prototype mirror and ex situ metrology, it is shown that the accuracy of our trained model enables open-loop shape control across a diverse set of states and that the control algorithm achieves shape error magnitudes that fall within diffraction-limited performance.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

TH-EF-BRB-02: Feasibility of Optimization for Dynamic Trajectory Radiotherapy

Purpose: Over the last years, volumetric modulated arc therapy (VMAT) has been widely introduced into clinical routine using a coplanar delivery technique. However, VMAT might be improved by including dynamic couch and collimator rotations, leading to dynamic trajectory radiotherapy (DTRT). In this work the feasibility and the potential benefit of DTRT was investigated. Methods: A general framework for the optimization was developed using the Eclipse Scripting Research Application Programming Interface (ESRAPI). Based on contoured target and organs at risk (OARs), the structures are extracted using the ESRAPI. Sampling potential beam directions, regularly distributed on a sphere using a Fibanocci-lattice, the fractional volume-overlap of each OAR and the target is determined and used to establish dynamic gantry-couch movements. Then, for each gantry-couch track the most suitable collimator angle is determined for each control point by optimizing the area between the MLC leaves and the target contour. The resulting dynamic trajectories are used as input to perform the optimization using a research version of the VMAT optimization algorithm and the ESRAPI. The feasibility of this procedure was tested for a clinically motivated head and neck case. Resulting dose distributions for the VMAT plan and for the dynamic trajectory treatment plan were compared based on DVH-parameters. Results: While the DVH for the target is virtually preserved, improvements in maximum dose for the DTRT plan were achieved for all OARs except for the inner-ear, where maximum dose remains the same. The major improvements in maximum dose were 6.5% of the prescribed dose (66 Gy) for the parotid and 5.5% for the myelon and the eye. Conclusion: The result of this work suggests that DTRT has a great potential to reduce dose to OARs with similar target coverage when compared to conventional VMAT treatment plans. This work was supported by Varian Medical Systems. This work was supported by Varian Medical Systems.

60 APPLIED LIFE SCIENCES↗

Inverse design of equilibrium cluster fluids applied to a physically informed model

Inverse design strategies have proven highly useful for the discovery of interaction potentials that prompt self-assembly of a variety of interesting structures. However, often the optimized particle interactions do not have a direct relationship to experimental systems. In this work, we show that Relative Entropy minimization is able to discover physically meaningful parameter sets for a model interaction built from depletion attraction and electrostatic repulsion that yield self-assembly of size-specific clusters. Here, we then explore the sensitivity of the optimized interaction potentials with respect to deviations in the underlying physical quantities, showing that clustering behavior is largely preserved even as the optimized parameters are perturbed.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

TROPHY: A Topologically Robust Physics-Informed Tracking Framework for Tropical Cyclones

Tropical cyclones (TCs) are among the most destructive weather systems. Realistically and efficiently detecting and tracking TCs are critical for assessing their impacts and risks. In particular, the eye is a signature feature of a mature TC. Therefore, knowing the eyes’ locations and movements is crucial for both operational weather forecasts and climate risk assessments. Recently, a multilevel robustness framework has been introduced to study the critical points of time-varying vector fields. The framework quantifies the robustness (i.e., structural stability) of critical points across varying neighborhoods. By relating the multilevel robustness with critical point tracking, the framework has demonstrated its potential in cyclone tracking. An advantage is that it identifies cyclonic features using only 2D wind vector fields, which is encouraging as most tracking algorithms require multiple dynamic and thermodynamic variables at different altitudes. A disadvantage is that the framework does not scale well computationally for datasets containing a large number of cyclones. Herein this paper introduces a topologically robust physics-informed tracking framework (TROPHY) for TC tracking. The main idea is to integrate physical knowledge of TC to drastically improve the computational efficiency of multilevel robustness framework for large-scale climate datasets. First, during preprocessing, we propose a physics-informed feature selection strategy to filter 90% of critical points that are short-lived and have low stability, thus preserving good candidates for TC tracking. Second, during in-processing, we impose constraints during the multilevel robustness computation to focus only on physics-informed neighborhoods of TCs. We apply TROPHY to 30 years of 2D wind fields from reanalysis data in ERA5 and generate a number of TC tracks. In comparison with the observed tracks, we demonstrate that TROPHY can capture TC characteristics (e.g., frequency, intensity, duration, latitudes with maximum intensity, and genesis) that are comparable to and sometimes even better than a well-validated TC tracking algorithm that requires multiple dynamic and thermodynamic scalar fields.

97 MATHEMATICS AND COMPUTING↗

Simultaneous cryo X-ray ptychographic and fluorescence microscopy of green algae

Trace metals play important roles in normal and in disease-causing biological functions. X-ray fluorescence microscopy reveals trace elements with no dependence on binding affinities (unlike with visible light fluorophores) and with improved sensitivity relative to electron probes. However, X-ray fluorescence is not very sensitive for showing the light elements that comprise the majority of cellular material. Here we show that X-ray ptychography can be combined with fluorescence to image both cellular structure and trace element distribution in frozen-hydrated cells at cryogenic temperatures, with high structural and chemical fidelity. Ptychographic reconstruction algorithms deliver phase and absorption contrast images at a resolution beyond that of the illuminating lens or beam size. Using 5.2-keV X-rays, we have obtained sub-30-nm resolution structural images and similar to 90-nm-resolution fluorescence images of several elements in frozen-hydrated green algae. This combined approach offers a way to study the role of trace elements in their structural context.

Deng, Junjing↗

Cross-scale Efficient Tensor Contractions for Coupled Cluster Computations Through Multiple Programming Model Backends

Coupled-cluster methods provide highly accurate models of molecular structure by explicit numerical calculation of tensors representing the correlation between electrons. These calculations are dominated by a sequence of tensor contractions, motivating the development of numerical libraries for such operations. While based on matrix-matrix multiplication, these libraries are specialized to exploit symmetries in the molecular structure and in electronic interactions, and thus reduce the size of the tensor representation and the complexity of contractions. The resulting algorithms are irregular and their parallelization has been previously achieved via the use of dynamic scheduling or specialized data decompositions. We introduce our efforts to extend the Libtensor framework to work in the distributed memory environment in a scalable and energy efficient manner. We achieve up to 240 speedup compared with the best optimized shared memory implementation. We attain scalability to hundreds of thousands of compute cores on three distributed-memory architectures, (Cray XC30&XC40, BlueGene/Q), and on a heterogeneous GPU-CPU system (Cray XK7). As the bottlenecks shift from being compute-bound DGEMM's to communication-bound collectives as the size of the molecular system scales, we adopt two radically different parallelization approaches for handling load-imbalance. Nevertheless, we preserve a uni ed interface to both programming models to maintain the productivity of computational quantum chemists.

74 ATOMIC AND MOLECULAR PHYSICS↗

WE-FG-207B-05: Iterative Reconstruction Via Prior Image Constrained Total Generalized Variation for Spectral CT

Purpose: To investigate iterative reconstruction via prior image constrained total generalized variation (PICTGV) for spectral computed tomography (CT) using fewer projections while achieving greater image quality. Methods: The proposed PICTGV method is formulated as an optimization problem, which balances the data fidelity and prior image constrained total generalized variation of reconstructed images in one framework. The PICTGV method is based on structure correlations among images in the energy domain and high-quality images to guide the reconstruction of energy-specific images. In PICTGV method, the high-quality image is reconstructed from all detector-collected X-ray signals and is referred as the broad-spectrum image. Distinct from the existing reconstruction methods applied on the images with first order derivative, the higher order derivative of the images is incorporated into the PICTGV method. An alternating optimization algorithm is used to minimize the PICTGV objective function. We evaluate the performance of PICTGV on noise and artifacts suppressing using phantom studies and compare the method with the conventional filtered back-projection method as well as TGV based method without prior image. Results: On the digital phantom, the proposed method outperforms the existing TGV method in terms of the noise reduction, artifacts suppression, and edge detail preservation. Compared to that obtained by the TGV based method without prior image, the relative root mean square error in the images reconstructed by the proposed method is reduced by over 20%. Conclusion: The authors propose an iterative reconstruction via prior image constrained total generalize variation for spectral CT. Also, we have developed an alternating optimization algorithm and numerically demonstrated the merits of our approach. Results show that the proposed PICTGV method outperforms the TGV method for spectral CT.

60 APPLIED LIFE SCIENCES↗

Two-tier model reduction of viscoelastically damped finite element models

Transient simulations of linear viscoelastically damped structures necessitate critical computational resources to directly integrate the full-order finite element model with time-stepping algorithms. Traditional modal reduction techniques are not directly applicable to these systems since viscoelastic materials depend on time and frequency. A more appropriate reduction basis is obtained from the nonlinear, complex eigenvalue problem, whose eigenvectors capture the appropriate kinematics and enable frequency-based mode selection; unfortunately, the computational cost is prohibitive for computing these modes from large-scale engineering models. To address this shortcoming, this work proposes a novel two-tier reduction procedure to reduce the upfront cost of solving the complex, nonlinear eigenvalue problem. The first reduction step reduces the full-order model with real mode shapes linearized about various centering frequencies to capture the kinematics over a full range of viscoelastic material behavior (glassy, rubbery, and glass-transition zones). This tier-one reduction preserves time-temperature superposition and allows the equations to depend parametrically on operating temperature. The second-level reduction then solves the complex, nonlinear eigenmode solutions in the tier-one reduced space about a fixed temperature to further reduce the equations-of-motion. The method is demonstrated on a cantilevered sandwich plate to display its accuracy and efficiency in comparison to full-order model predictions.

97 MATHEMATICS AND COMPUTING↗

TU-AB-202-05: GPU-Based 4D Deformable Image Registration Using Adaptive Tetrahedral Mesh Modeling

Purpose: Deformable image registration (DIR) has been employed today as an automated and effective segmentation method to transfer tumor or organ contours from the planning image to daily images, instead of manual segmentation. However, the computational time and accuracy of current DIR approaches are still insufficient for online adaptive radiation therapy (ART), which requires real-time and high-quality image segmentation, especially in a large datasets of 4D-CT images. The objective of this work is to propose a new DIR algorithm, with fast computational speed and high accuracy, by using adaptive feature-based tetrahedral meshing and GPU-based parallelization. Methods: The first step is to generate the adaptive tetrahedral mesh based on the image features of a reference phase of 4D-CT, so that the deformation can be well captured and accurately diffused from the mesh vertices to voxels of the image volume. Subsequently, the deformation vector fields (DVF) and other phases of 4D-CT can be obtained by matching each phase of the target 4D-CT images with the corresponding deformed reference phase. The proposed 4D DIR method is implemented on GPU, resulting in significantly increasing the computational efficiency due to its parallel computing ability. Results: A 4D NCAT digital phantom was used to test the efficiency and accuracy of our method. Both the image and DVF results show that the fine structures and shapes of lung are well preserved, and the tumor position is well captured, i.e., 3D distance error is 1.14 mm. Compared to the previous voxel-based CPU implementation of DIR, such as demons, the proposed method is about 160x faster for registering a 10-phase 4D-CT with a phase dimension of 256×256×150. Conclusion: The proposed 4D DIR method uses feature-based mesh and GPU-based parallelism, which demonstrates the capability to compute both high-quality image and motion results, with significant improvement on the computational speed.

60 APPLIED LIFE SCIENCES↗

A Measure-Theoretic Algorithm for Estimating Bottom Friction in a Coastal Inlet: Case Study of Bay St. Louis during Hurricane Gustav (2008)

The majority of structural damage and loss of life during a hurricane is due to storm surge, thus it is important for communities in hurricane-prone regions to understand their risk due to surge. Storm surge in particular is largely influenced by coastal features such as topography/bathymetry and bottom roughness. Bottom roughness determines how much resistance there is to the flow. Manning’s formula can be used to model the bottom stress with the Manning’s n coefficient, a spatially dependent field. Given a storm surge model and a set of model outputs, an inverse problem may be solved to determine probable Manning’s n fields to use for predictive simulations. The inverse problem is formulated and solved in a measure-theoretic framework using the state-of-the-art Advanced Circulation (ADCIRC) storm surge model. The use of measure theory requires minimal assumptions and involves the direct inversion of the physics-based map from model inputs to output data determined by the ADCIRC model. Thus, key geometric relationships in this map are preserved and exploited. By using a recently available subdomain implementation of ADCIRC that significantly reduces the computational cost of forward model solves, the authors demonstrate the method on a case study using data obtained from an ADCIRC hindcast study of Hurricane Gustav (2008) to quantify uncertainties in Manning’s n within Bay St. Louis. However, the methodology is general and could be applied to any inverse problem that involves a map from model input to output quantities of interest.

54 ENVIRONMENTAL SCIENCES↗

Cross-scale efficient tensor contractions for coupled cluster computations through multiple programming model backends

Coupled-cluster methods provide highly accurate models of molecular structure through explicit numerical calculation of tensors representing the correlation between electrons. These calculations are dominated by a sequence of tensor contractions, motivating the development of numerical libraries for such operations. While based on matrix–matrix multiplication, these libraries are specialized to exploit symmetries in the molecular structure and in electronic interactions, and thus reduce the size of the tensor representation and the complexity of contractions. The resulting algorithms are irregular and their parallelization has been previously achieved via the use of dynamic scheduling or specialized data decompositions. We introduce our efforts to extend the Libtensor framework to work in the distributed memory environment in a scalable and energy-efficient manner. We achieve up to 240× speedup compared with the optimized shared memory implementation of Libtensor. We attain scalability to hundreds of thousands of compute cores on three distributed-memory architectures (Cray XC30 and XC40, and IBM Blue Gene/Q), and on a heterogeneous GPU-CPU system (Cray XK7). As the bottlenecks shift from being compute-bound DGEMM's to communication-bound collectives as the size of the molecular system scales, we adopt two radically different parallelization approaches for handling load-imbalance, tasking and bulk synchronous models. Nevertheless, we preserve a unified interface to both programming models to maintain the productivity of computational quantum chemists.

97 MATHEMATICS AND COMPUTING↗

Ultrafast processing of pixel detector data with machine learning frameworks

Modern photon science performed at high repetition rate free-electron laser (FEL) facilities and beyond relies on 2D pixel detectors operating at increasing frequencies (towards 100 kHz at LCLS-II) and producing rapidly increasing amounts of data (towards TB/s). This data must be rapidly stored for offline analysis and summarized in real time for online feedback to the scientists. While at LCLS all raw data has been stored, at LCLS-II this would lead to a prohibitive cost; instead, enabling real time processing of pixel detector data (dark, gain, common mode, background, charge summing, subpixel position, photon counting, data summarization) allows reducing the size and cost of online processing, offline processing and storage by orders of magnitude while preserving full photon information. This could be achieved by taking advantage of the compressibility of sparse data typical for LCLS-II applications. Faced with a similar big data challenge a decade ago, computer vision stimulated revolutionary advances in machine learning hardware and software. We investigated if these developments are useful in the field of data processing for high speed pixel detectors and found that typical deep learning models and autoencoder architectures failed to yield useful noise reduction while preserving full photon information, presumably because of the very different statistics and feature sets in computer vision and radiation imaging. However, the raw performance of modern frameworks like Tensorflow inspired us to redesign in Tensorflow mathematically equivalent versions of the state-of-the-art, “classical” algorithms used at LCLS. The novel Tensorflow models resulted in elegant, compact and hardware agnostic code, gaining 1 to 2 orders of magnitude faster processing on an inexpensive consumer GPU, reducing by 3 orders of magnitude the projected cost of online analysis and compression without photon loss at LCLS-II. The novel Tensorflow models also enabled ongoing development of a pipelined hardware system expected to yield an additional 3 to 4 orders of magnitude speedup, necessary for meeting the data acquisition and storage requirements at LCLS-II, potentially enabling acquiring every single FEL pulse at full speed. Computer vision a decade ago was dominated by hand-crafted filters; their structure inspired the deep learning revolution resulting in modern deep convolutional networks; similarly, our novel Tensorflow filters provide inspiration for designing future deep learning models for ultrafast and efficient processing and classification of pixel detector images at FEL facilities.

47 OTHER INSTRUMENTATION↗