Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “distributed 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 469 records · Page 26
All sourcesNASA NTRSDOE OSTI
DOE OSTI2014

Helium segregation to screw and edge dislocations in α-iron and their yield strength

Helium (He) introduces one of the mayor concerns in the nuclear materials community as it modifies the mechanical properties of the system withstanding fast neutron spectra, promoting swelling and embrittlement. Ferritic/martensitic steels are one of the main candidates as structural materials for future nuclear applications. Experimentally the bubble distribution is observed to vary depending on irradiation conditions (temperature, dose rate and total dose). However, traditional atomistic models decouple the role of temperature in the mechanical properties from its effect on the bubble distribution. In this paper we study substitutional He segregation to screw and edge dislocations in α-Fe at different temperatures. We use an object kinetic Monte Carlo methodology to obtain general trends in bubble distribution and a canonical Monte Carlo algorithm, with full atomistic fidelity, to find the He distribution at the dislocation cores. Molecular dynamics has subsequently been applied to study the yield strength, which increases significantly in the presence of He, more remarkably for the edge dislocation. The total stress fits a Kocks relation. Yet, if the lattice resistance is subtracted, the relation between the critical shear stress and the temperature is non-monotonic for the screw character. To reproduce this effect, we propose to modify the Kocks relation, adding a second-order term in temperature that extends the range of applicability of the model.

36 MATERIALS SCIENCE↗
DOE OSTI2015

SU-E-T-456: Impact of Dose Calculation Algorithms On Biologically Optimized VMAT Plans for Esophageal Cancer

Purpose: To evaluate the impact of dose calculation algorithm on the dose distribution of biologically optimized Volumatric Modulated Arc Therapy (VMAT) plans for Esophgeal cancer. Methods: Eighteen retrospectively treated patients with carcinoma esophagus were studied. VMAT plans were optimized using biological objectives in Monaco (5.0) TPS for 6MV photon beam (Elekta Infinity). These plans were calculated for final dose using Monte Carlo (MC), Collapsed Cone Convolution (CCC) & Pencil Beam Convolution (PBC) algorithms from Monaco and Oncentra Masterplan TPS. A dose grid of 2mm was used for all algorithms and 1% per plan uncertainty maintained for MC calculation. MC based calculations were considered as the reference for CCC & PBC. Dose volume histogram (DVH) indices (D95, D98, D50 etc) of Target (PTV) and critical structures were compared to study the impact of all three algorithms. Results: Beam models were consistent with measured data. The mean difference observed in reference with MC calculation for D98, D95, D50 & D2 of PTV were 0.37%, −0.21%, 1.51% & 1.18% respectively for CCC and 3.28%, 2.75%, 3.61% & 3.08% for PBC. Heart D25 mean difference was 4.94% & 11.21% for CCC and PBC respectively. Lung Dmean mean difference was 1.5% (CCC) and 4.1% (PBC). Spinal cord D2 mean difference was 2.35% (CCC) and 3.98% (PBC). Similar differences were observed for liver and kidneys. The overall mean difference found for target and critical structures was 0.71±1.52%, 2.71±3.10% for CCC and 3.18±1.55%, 6.61±5.1% for PBC respectively. Conclusion: We observed a significant overestimate of dose distribution by CCC and PBC as compared to MC. The dose prediction of CCC is closer (<3%) to MC than that of PBC. This can be attributed to poor performance of CCC and PBC in inhomogeneous regions around esophagus. CCC can be considered as an alternate in the absence of MC algorithm.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

Initializing BSQ with Open-Source ICCING

While it is well known that there is a significant amount of conserved charges in the initial state of nuclear collisions, the production of these due to gluon splitting has yet to be thoroughly investigated. The ICCING (Initial Conserved Charges in Nuclear Geometry) algorithm reconstructs these quark distributions, providing conserved strange, baryon, and electric charges, by sampling a given model for the g → qq¯ splitting function over the initial energy density, which is valid at top collider energies, even when µB = 0. The ICCING algorithm includes fluctuations in the gluon longitudinal momenta, a structure that supports the implementation of dynamical processes, and the c++ version is now open-source. A full analysis of parameter choices on the model has been done to quantify the effect these have on the underlying physics. We find there is a sustained difference across the different charges that indicates sensitivity to hot spot geometry.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2023

An Initial Study of the Convergence Rate of Griffin’s Pebble Bed Reactors Algorithm

This paper presents an initial study of the convergence properties of an iterative algorithm for computing the burnup distribution in a pebble bed reactor (PBR) in its equilibrium core condition. The algorithm is implemented in the Griffin code. Griffin is a reactor multiphysics analysis application jointly developed by Idaho National Laboratory (INL) and Argonne National Laboratory (ANL). Griffin’s PBR algorithm is discussed and simulation data are presented. An alternative matrix formulation of the algorithm is presented that facilitates analysis of the iterative algorithm. The dependence of the spectral radius of the iterative algorithm on operational and discretization parameters is investigated.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

DESI mock challenge: Halo and galaxy catalogues with the bias assignment method

We present a novel approach to the construction of mock galaxy catalogues for large-scale structure analysis based on the distribution of dark matter halos obtained with effective bias models at the field level. We aim to produce mock galaxy catalogues capable of generating accurate covariance matrices for a number of cosmological probes that are expected to be measured in current and forthcoming galaxy redshift surveys (e.g. two- and three-point statistics). The construction of the catalogues shown in this paper is part of a mock-comparison project within the Dark Energy Spectroscopic Instrument (DESI) collaboration. We use the bias assignment method ( BAM ) to model the statistics of halo distribution through a learning algorithm using a few detailed N-body simulations, and approximated gravity solvers based on Lagrangian perturbation theory. We introduce cosmic-web-dependent corrections to modelling redshift-space distortions at the N-body level – both in the halo and galaxy distributions –, as well as a multi-scale approach for accurate assignment of halo properties. Using specific models of halo occupation distributions to populate halos, we generate galaxy mocks with the expected number density and central-satellite fraction of emission-line galaxies, which are a key target of the DESI experiment. BAM generates mock catalogues with per cent accuracy in a number of summary statistics, such as the abundance, the two- and three-point statistics of halo distributions, both in real and redshift space. In particular, the mock galaxy catalogues display ~3%-10% accuracy in the multipoles of the power spectrum up to scales of k ~ 0.4 h -1 Mpc. We show that covariance matrices of two- and three-point statistics obtained with BAM display a similar structure to the reference simulation. BAM offers an efficient way to produce mock halo catalogues with accurate two- and three-point statistics and is able to generate a variety of multi-tracer catalogues with precise covariance matrices of several cosmological probes. We discuss future developments of the algorithm towards mock production in DESI and other galaxy-redshift surveys.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2016

SU-D-204-07: Retrospective Correlation of Dose Accuracy with Regions of Local Failure for Early Stage Lung Cancer Patients Treated with Stereotactic Body Radiotherapy

Purpose: To correlate dose distributions computed using six algorithms for recurrent early stage non-small cell lung cancer (NSCLC) patients treated with stereotactic body radiotherapy (SBRT), with outcome (local failure). Methods: Of 270 NSCLC patients treated with 12Gyx4, 20 were found to have local recurrence prior to the 2-year time point. These patients were originally planned with 1-D pencil beam (1-D PB) algorithm. 4D imaging was performed to manage tumor motion. Regions of local failures were determined from follow-up PET-CT scans. Follow-up CT images were rigidly fused to the planning CT (pCT), and recurrent tumor volumes (Vrecur) were mapped to the pCT. Dose was recomputed, retrospectively, using five algorithms: 3-D PB, collapsed cone convolution (CCC), anisotropic analytical algorithm (AAA), AcurosXB, and Monte Carlo (MC). Tumor control probability (TCP) was computed using the Marsden model (1,2). Patterns of failure were classified as central, in-field, marginal, and distant for Vrecur ≥95% of prescribed dose, 95–80%, 80–20%, and ≤20%, respectively (3). Results: Average PTV D95 (dose covering 95% of the PTV) for 3-D PB, CCC, AAA, AcurosXB, and MC relative to 1-D PB were 95.3±2.1%, 84.1±7.5%, 84.9±5.7%, 86.3±6.0%, and 85.1±7.0%, respectively. TCP values for 1-D PB, 3-D PB, CCC, AAA, AcurosXB, and MC were 98.5±1.2%, 95.7±3.0, 79.6±16.1%, 79.7±16.5%, 81.1±17.5%, and 78.1±20%, respectively. Patterns of local failures were similar for 1-D and 3D PB plans, which predicted that the majority of failures occur in centraldistal regions, with only ∼15% occurring distantly. However, with convolution/superposition and MC type algorithms, the majority of failures (65%) were predicted to be distant, consistent with the literature. Conclusion: Based on MC and convolution/superposition type algorithms, average PTV D95 and TCP were ∼15% lower than the planned 1-D PB dose calculation. Patterns of failure results suggest that MC and convolution/superposition type algorithms predict different outcomes for patterns of failure relative to PB algorithms. Work supported in part by Varian Medical Systems, Palo Alto, CA.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

Real-Time Hardware-in-the-Loop Testbed to Evaluate FLISR Implemented with OpenFMB

With the increasing complexity of the distribution smart grid architecture, algorithms such as the fault location, isolation, and service restoration (FLISR) scheme rely on robust communications that are resilient to natural and man-made adverse conditions and exhibit robustness. Existing communications infrastructure for information exchange are centralized at the distribution management system, with very little autonomy or intelligence at the grid-edge. As a first step towards achieving grid-edge self-healing, this paper aims to bridge this shortcoming by implementing a centrally coordinated rules-based FLISR scheme and integrating it with Open Field Message Bus (OpenFMB), which is a flexible publish-subscribe architecture with the potential to enable point-to-multipoint communications and is more robust and resilient to natural and man-made adverse conditions. A proof of concept is developed to validate the centrally coordinated FLISR and OpenFMB mounted on an SEL-3360 computer that interacts with a simple feeder network of five SEL-651R relays, an SEL-3530 RTAC, and a hardware-in-the-loop testbed. Results demonstrate the efficacy of this approach in enabling direct, low-latency information exchange. OpenFMB's publish-subscribe data model also opens new ways to enable grid-edge interoperability among devices of different vendors interacting with different protocols.

Sundararajan, Aditya↗
DOE OSTI2018

Mathematical Model for the Emission Infrared Tomography of the Temperature Field in an Isotropic Layer

We consider a mathematical model for the determination of the temperature field in a layer that emits, absorbs, and scatters infrared radiation both in the volume and on the surface. Within the framework of the model, we formulate nonlinear direct and inverse problems of emission infrared tomography of the temperature field in the layer according to known space-and-angular distributions of the flows of infrared radiation emitted into the ambient medium. The iterative-variational methods are proposed for the solution of the indicated direct and inverse problems. As a specific example, we perform the numerical analysis of the developed algorithms.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2016

Assessing the pollution risk of a groundwater source field at western Laizhou Bay under seawater intrusion

Coastal areas have great significance for human living, economy and society development in the world. With the rapid increase of pressures from human activities and climate change, the safety of groundwater resource is under the threat of seawater intrusion in coastal areas. The area of Laizhou Bay is one of the most serious seawater intruded areas in China, since seawater intrusion phenomenon was firstly recognized in the middle of 1970s. This study assessed the pollution risk of a groundwater source filed of western Laizhou Bay area by inferring the probability distribution of groundwater Cl{sup −} concentration. The numerical model of seawater intrusion process is built by using SEAWAT4. The parameter uncertainty of this model is evaluated by Markov Chain Monte Carlo (MCMC) simulation, and DREAM{sub (ZS)} is used as sampling algorithm. Then, the predictive distribution of Cl{sup -} concentration at groundwater source field is inferred by using the samples of model parameters obtained from MCMC. After that, the pollution risk of groundwater source filed is assessed by the predictive quantiles of Cl{sup -} concentration. The results of model calibration and verification demonstrate that the DREAM{sub (ZS)} based MCMC is efficient and reliable to estimate model parameters under current observation. Under the condition of 95% confidence level, the groundwater source point will not be polluted by seawater intrusion in future five years (2015–2019). In addition, the 2.5% and 97.5% predictive quantiles show that the Cl{sup −} concentration of groundwater source field always vary between 175 mg/l and 200 mg/l. - Highlights: • The parameter uncertainty of seawater intrusion model is evaluated by MCMC. • Groundwater source field won’t be polluted by seawater intrusion in future 5 years. • The pollution risk is assessed by the predictive quantiles of Cl{sup −} concentration.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2016

Dynamical Mass Measurements of Contaminated Galaxy Clusters Using Machine Learning

We study dynamical mass measurements of galaxy clusters contaminated by interlopers and show that a modern machine learning algorithm can predict masses by better than a factor of two compared to a standard scaling relation approach. We create two mock catalogs from Multidark's publicly available N-body MDPL1 simulation, one with perfect galaxy cluster membership information and the other where a simple cylindrical cut around the cluster center allows interlopers to contaminate the clusters. In the standard approach, we use a power-law scaling relation to infer cluster mass from galaxy line-of-sight (LOS) velocity dispersion. Assuming perfect membership knowledge, this unrealistic case produces a wide fractional mass error distribution, with a width of ${\rm{\Delta }}\epsilon \approx 0.87$. Interlopers introduce additional scatter, significantly widening the error distribution further (${\rm{\Delta }}\epsilon \approx 2.13$). We employ the support distribution machine (SDM) class of algorithms to learn from distributions of data to predict single values. Applied to distributions of galaxy observables such as LOS velocity and projected distance from the cluster center, SDM yields better than a factor-of-two improvement (${\rm{\Delta }}\epsilon \approx 0.67$) for the contaminated case. Remarkably, SDM applied to contaminated clusters is better able to recover masses than even the scaling relation approach applied to uncontaminated clusters. We show that the SDM method more accurately reproduces the cluster mass function, making it a valuable tool for employing cluster observations to evaluate cosmological models.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2015

Influence of source batch S{sub K} dispersion on dosimetry for prostate cancer treatment with permanent implants

Purpose: In clinical practice, specific air kerma strength (S{sub K}) value is used in treatment planning system (TPS) permanent brachytherapy implant calculations with {sup 125}I and {sup 103}Pd sources; in fact, commercial TPS provide only one S{sub K} input value for all implanted sources and the certified shipment average is typically used. However, the value for S{sub K} is dispersed: this dispersion is not only due to the manufacturing process and variation between different source batches but also due to the classification of sources into different classes according to their S{sub K} values. The purpose of this work is to examine the impact of S{sub K} dispersion on typical implant parameters that are used to evaluate the dose volume histogram (DVH) for both planning target volume (PTV) and organs at risk (OARs). Methods: The authors have developed a new algorithm to compute dose distributions with different S{sub K} values for each source. Three different prostate volumes (20, 30, and 40 cm{sup 3}) were considered and two typical commercial sources of different radionuclides were used. Using a conventional TPS, clinically accepted calculations were made for {sup 125}I sources; for the palladium, typical implants were simulated. To assess the many different possible S{sub K} values for each source belonging to a class, the authors assigned an S{sub K} value to each source in a randomized process 1000 times for each source and volume. All the dose distributions generated for each set of simulations were assessed through the DVH distributions comparing with dose distributions obtained using a uniform S{sub K} value for all the implanted sources. The authors analyzed several dose coverage (V{sub 100} and D{sub 90}) and overdosage parameters for prostate and PTV and also the limiting and overdosage parameters for OARs, urethra and rectum. Results: The parameters analyzed followed a Gaussian distribution for the entire set of computed dosimetries. PTV and prostate V{sub 100} and D{sub 90} variations ranged between 0.2% and 1.78% for both sources. Variations for the overdosage parameters V{sub 150} and V{sub 200} compared to dose coverage parameters were observed and, in general, variations were larger for parameters related to {sup 125}I sources than {sup 103}Pd sources. For OAR dosimetry, variations with respect to the reference D{sub 0.1cm{sup 3}} were observed for rectum values, ranging from 2% to 3%, compared with urethra values, which ranged from 1% to 2%. Conclusions: Dose coverage for prostate and PTV was practically unaffected by S{sub K} dispersion, as was the maximum dose deposited in the urethra due to the implant technique geometry. However, the authors observed larger variations for the PTV V{sub 150}, rectum V{sub 100}, and rectum D{sub 0.1cm{sup 3}} values. The variations in rectum parameters were caused by the specific location of sources with S{sub K} value that differed from the average in the vicinity. Finally, on comparing the two sources, variations were larger for {sup 125}I than for {sup 103}Pd. This is because for {sup 103}Pd, a greater number of sources were used to obtain a valid dose distribution than for {sup 125}I, resulting in a lower variation for each S{sub K} value for each source (because the variations become averaged out statistically speaking)

60 APPLIED LIFE SCIENCES↗
DOE OSTI2012

A pencil beam algorithm for helium ion beam therapy

Purpose: To develop a flexible pencil beam algorithm for helium ion beam therapy. Dose distributions were calculated using the newly developed pencil beam algorithm and validated using Monte Carlo (MC) methods. Methods: The algorithm was based on the established theory of fluence weighted elemental pencil beam (PB) kernels. Using a new real-time splitting approach, a minimization routine selects the optimal shape for each sub-beam. Dose depositions along the beam path were determined using a look-up table (LUT). Data for LUT generation were derived from MC simulations in water using GATE 6.1. For materials other than water, dose depositions were calculated by the algorithm using water-equivalent depth scaling. Lateral beam spreading caused by multiple scattering has been accounted for by implementing a non-local scattering formula developed by Gottschalk. A new nuclear correction was modelled using a Voigt function and implemented by a LUT approach. Validation simulations have been performed using a phantom filled with homogeneous materials or heterogeneous slabs of up to 3 cm. The beams were incident perpendicular to the phantoms surface with initial particle energies ranging from 50 to 250 MeV/A with a total number of 10{sup 7} ions per beam. For comparison a special evaluation software was developed calculating the gamma indices for dose distributions. Results: In homogeneous phantoms, maximum range deviations between PB and MC of less than 1.1% and differences in the width of the distal energy falloff of the Bragg-Peak from 80% to 20% of less than 0.1 mm were found. Heterogeneous phantoms using layered slabs satisfied a {gamma}-index criterion of 2%/2mm of the local value except for some single voxels. For more complex phantoms using laterally arranged bone-air slabs, the {gamma}-index criterion was exceeded in some areas giving a maximum {gamma}-index of 1.75 and 4.9% of the voxels showed {gamma}-index values larger than one. The calculation precision of the presented algorithm was considered to be sufficient for clinical practice. Although only data for helium beams was presented, the performance of the pencil beam algorithm for proton beams was comparable. Conclusions: The pencil beam algorithm developed for helium ions presents a suitable tool for dose calculations. Its calculation speed was evaluated to be similar to other published pencil beam algorithms. The flexible design allows easy customization of measured depth-dose distributions and use of varying beam profiles, thus making it a promising candidate for integration into future treatment planning systems. Current work in progress deals with RBE effects of helium ions to complete the model.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2017

Distributed Control of Inverter-Interfaced Microgrids Based on Consensus Algorithm with Improved Transient Performance

Conventional control solutions of the inverter-interfaced microgrids are usually designed based on models with fully decoupled subsystems. The negligence of the strong coupling due to power lines impedance leads to large transient line currents, which might trigger false protection. Besides, the droop-based control methods unnecessarily introduce system frequency and voltage deviations. To overcome these issues, a novel distributed control scheme is proposed for the inverter-interfaced microgrids in this study. The objective of the primary control is to regulate the bus voltages and frequency while suppressing the transient line currents. The objective of secondary control is to maintain fair load sharing. Both primary control and secondary control are distributed and subsystems or control agents only require measurements from local and neighboring subsystems. The detailed control problem formulation, control design and stability analysis are presented in the paper. Finally, the effectiveness of the proposed control solution is evaluated through extensive simulations based on both simplified and detailed models.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2019

A communication-avoiding 3D algorithm for sparse LU factorization on heterogeneous systems

We propose a new algorithm to improve the strong scalability of right-looking sparse LU factorization on distributed memory systems. Our 3D algorithm for sparse LU uses a three-dimensional MPI process grid, exploits elimination tree parallelism, and trades off increased memory for reduced per-process communication. We also analyze the asymptotic improvements for planar graphs (e.g., those arising from 2D grid or mesh discretizations) and certain non-planar graphs (specifically for 3D grids and meshes). For a planar graph with $n$ vertices, our algorithm reduces communication volume asymptotically in $n$ by a factor of $\mathscr{O}$ $\Big(\sqrt{log \ n}\Big)$ and latency by a factor of $\mathscr{O}$ $(log \ n)$. For nonplanar cases, our algorithm can reduce the per-process communication volume by 3× and latency by $\mathscr{O}$ $\Big(n^\frac13\Big)$ times. In all cases, the memory needed to achieve these gains is a constant factor. We implemented our algorithm by extending the 2D data structure used in SuperLU_DIST. Our new 3D code achieves empirical speedups up to 27× for planar graphs and up to 3.3× for non-planar graphs over the baseline 2D SuperLU_DIST when run on 24,000 cores of a Cray XC30. We extend the 3D algorithm for heterogeneous architectures by adding the Highly Asynchronous Lazy Offload (Halo) algorithm for co-processor offload. On 4096 nodes of a Cray XK7 with 32,768 CPU cores and 4096 Nvidia K20x GPUs, the 3D algorithm achieves empirical speedups up to 24× for planar graphs and 3.5× for non-planar graphs over the baseline 2D SuperLU_DIST with co-processor acceleration.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2013

Expectation-maximization algorithms for learning a finite mixture of univariate survival time distributions from partially specified class values

Heterogeneity exists on a data set when samples from di erent classes are merged into the data set. Finite mixture models can be used to represent a survival time distribution on heterogeneous patient group by the proportions of each class and by the survival time distribution within each class as well. The heterogeneous data set cannot be explicitly decomposed to homogeneous subgroups unless all the samples are precisely labeled by their origin classes; such impossibility of decomposition is a barrier to overcome for estimating nite mixture models. The expectation-maximization (EM) algorithm has been used to obtain maximum likelihood estimates of nite mixture models by soft-decomposition of heterogeneous samples without labels for a subset or the entire set of data. In medical surveillance databases we can find partially labeled data, that is, while not completely unlabeled there is only imprecise information about class values. In this study we propose new EM algorithms that take advantages of using such partial labels, and thus incorporate more information than traditional EM algorithms. We particularly propose four variants of the EM algorithm named EM-OCML, EM-PCML, EM-HCML and EM-CPCML, each of which assumes a specific mechanism of missing class values. We conducted a simulation study on exponential survival trees with five classes and showed that the advantages of incorporating substantial amount of partially labeled data can be highly signi cant. We also showed model selection based on AIC values fairly works to select the best proposed algorithm on each specific data set. A case study on a real-world data set of gastric cancer provided by Surveillance, Epidemiology and End Results (SEER) program showed a superiority of EM-CPCML to not only the other proposed EM algorithms but also conventional supervised, unsupervised and semi-supervised learning algorithms.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

TriC: Distributed-memory Triangle Counting by Exploiting the Graph Structure

Graph analytics has emerged as an important tool in the analysis of large scale data from diverse application domains such as social networks, cyber security and bioinformatics. Counting the number of triangles in a graph is a fundamental kernel with several applications such as detecting the community structure of a graph or in identifying important vertices in a graph. The ubiquity of massive datasets is driving the need to scale graph analytics on parallel systems. However, numerous challenges exist in efficiently parallelizing graph algorithms, especially on distributed-memory systems. Irregular memory accesses and communication patterns, low computation to communication ratios, and the need for frequent synchronization are some of the leading challenges. In this paper, we present TriC, our distributed-memory implementation of triangle counting in graphs using the Message Passing Interface (MPI), as a submission to the 2020 GraphChallenge competition. Using a set of synthetic and real-world inputs from the challenge, we demonstrate a speedup of up to 90x relative to previous work on 32 processor-cores of a NERSC Cori node. We also provide details from distributed runs with up to8192 processes along with strong scaling results. The observations presented in this work provide an understanding of the system-level bottlenecks at scale that specifically impact sparse-irregular workloads and will therefore benefit other efforts to parallelize graph algorithms.

Halappanavar, Mahantesh↗
DOE OSTI2014

EM reconstruction of dual isotope PET using staggered injections and prompt gamma positron emitters

Purpose: The aim of dual isotope positron emission tomography (DIPET) is to create two separate images of two coinjected PET radiotracers. DIPET shortens the duration of the study, reduces patient discomfort, and produces perfectly coregistered images compared to the case when two radiotracers would be imaged independently (sequential PET studies). Reconstruction of data from such simultaneous acquisition of two PET radiotracers is difficult because positron decay of any isotope creates only 511 keV photons; therefore, the isotopes cannot be differentiated based on the detected energy. Methods: Recently, the authors have proposed a DIPET technique that uses a combination of radiotracer A which is a pure positron emitter (such as{sup 18}F or {sup 11}C) and radiotracer B in which positron decay is accompanied by the emission of a high-energy (HE) prompt gamma (such as {sup 38}K or {sup 60}Cu). Events that are detected as triple coincidences of HE gammas with the corresponding two 511 keV photons allow the authors to identify the lines-of-response (LORs) of isotope B. These LORs are used to separate the two intertwined distributions, using a dedicated image reconstruction algorithm. In this work the authors propose a new version of the DIPET EM-based reconstruction algorithm that allows the authors to include an additional, independent estimate of radiotracer A distribution which may be obtained if radioisotopes are administered using a staggered injections method. In this work the method is tested on simple simulations of static PET acquisitions. Results: The authors’ experiments performed using Monte-Carlo simulations with static acquisitions demonstrate that the combined method provides better results (crosstalk errors decrease by up to 50%) than the positron-gamma DIPET method or staggered injections alone. Conclusions: The authors demonstrate that the authors’ new EM algorithm which combines information from triple coincidences with prompt gammas and staggered injections improves the accuracy of DIPET reconstructions for static acquisitions so they reach almost the benchmark level calculated for perfectly separated tracers.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2021

Explainable Neural Architecture Search (XNAS)

Code for the paper Learning Interpretable Models Through Multi-Objective Neural Architecture Search by Zachariah Carmichael, Tim Moon, and Sam Ade Jacobs. Monumental advances in deep learning have led to unprecedented achievements across a multitude of domains. While the performance of deep neural networks is indubitable, the architectural design and interpretability of such models are nontrivial. Research has been introduced to automate the design of neural network architectures through neural architecture search (NAS). Recent progress has made these methods more pragmatic by exploiting distributed computation and novel optimization algorithms. However, there is little work in optimizing architectures for interpretability. To this end, we propose a multiobjective distributed NAS framework that optimizes for both task performance and introspection. We leverage the non-dominated sorting genetic algorithm (NSGA-II) and explainable AI (XAI) techniques to reward architectures that can be better comprehended by humans. The framework is evaluated on several image classification datasets. We demonstrate that jointly optimizing for introspection ability and task error leads to more disentangled architectures that perform within tolerable error.

Carmichael, ZachariahJ↗