Search

Large scale, inverse problem solving deep learning algorithms have become an essential part of modern research and industrial applications. The complexity of the underlying inverse problem often poses challenges to the algorithm and requires the proper utilization of high-performance computing systems. Most deep learning algorithms require, due to their design, custom parallelization techniques in order to be resource efficient while showing a reasonable convergence. In this paper we introduces a \underline{S}calable \underline{A}synchronous \underline{G}enerative workflow for solving \underline{I}nverse \underline{P}roblems \underline{S}olver (SAGIPS) on high-performance computing systems. We present a workflow that utilizes a parallelization approach where the gradients of the generator network are updated in an asynchronous ring-all-reduce fashion. Experiments with a scientific proxy application demonstrate that SAGIPS shows near linear weak scaling, together with a convergence quality that is comparable to traditional methods. The approach presented here allows leveraging GANs across multiple GPUs, promising advancements in solving complex inverse problems at scale.

Massive upgrades to science infrastructure are driving data velocities upwards while stimulating adoption of increasingly data-intensive analytics. While next-generation exascale supercomputers promise strong support for I/O-intensive workflows, HPC remains largely untapped by live experiments, because data transfers and disparate batch-queueing policies are prohibitive when faced with scarce instrument time. To bridge this divide, we introduce Balsam: a distributed orchestration platform enabling workflows at the edge to securely and efficiently trigger analytics tasks across a user-managed federation of HPC execution sites. We describe the architecture of the Balsam service, which provides a workflow management API, and distributed sites that provision resources and schedule scalable, fault-tolerant execution. We demonstrate Balsam in efficiently scaling real-time analytics from two DOE light sources simultaneously onto three supercomputers (Theta, Summit, and Cori), while maintaining low overheads for on-demand computing, and providing a Python library for seamless integration with existing ecosystems of data analysis tools.

We introduce the Balsam service to manage high-throughput task scheduling and execution on supercomputing systems. Balsam allows users to populate a task database with a variety of tasks ranging from simple independent tasks to dynamic multi-task workflows. With abstractions for the local resource scheduler and MPI environment, Balsam dynamically packages tasks into ensemble jobs and manages their scheduling lifecycle. The ensembles execute in a pilot "launcher" which (i) ensures concurrent, load-balanced execution of arbitrary serial and parallel programs with heterogeneous processor requirements, (ii) requires no modification of user applications, (iii) is tolerant of task-level faults and provides several options for error recovery, (iv) stores provenance data (e.g task history, error logs) in the database, (v) supports dynamic workflows, in which tasks are created or killed at runtime. Here, we present the design and Python implementation of the Balsam service and launcher. The efficacy of this system is illustrated using two case studies: hyperparameter optimization of deep neural networks, and high-throughput single-point quantum chemistry calculations. We find that the unique combination of flexible job-packing and automated scheduling with dynamic (pilot-managed) execution facilitates excellent resource utilization. The scripting overheads typically needed to manage resources and launch workflows on supercomputers are substantially reduced, accelerating workflow development and execution., Comment: SC '18: 8th Workshop on Python for High-Performance and Scientific Computing (PyHPC 2018)

The areal centroids of the youngest polar deposits on Mars are offset from those of adjacent paleopolar deposits by 5-10 degrees. We test the hypothesis that the offset is the result of true polar wander (TPW), the motion of the solid surface with respect to the spin axis, caused by a mass redistribution within or on the surface of Mars. In particular, we consider TPW driven by late-stage volcanism during the late Hesperian to Amazonian. There is observational and qualitative support for this hypothesis: in both North and South, observed offsets lie close to a great circle 90 degrees from Tharsis, as expected for polar wander after Tharsis formed. We calculate the magnitude and direction of TPW produced by mapped late-stage lavas for a range of lithospheric thicknesses, lava thicknesses, eruption histories, and prior polar wander events. If Tharsis formed close to the equator, the stabilizing effect of a fossil rotational bulge located close to the equator leads to predicted TPW of <2 degrees, too small to account for observed offsets. If, however, Tharsis formed far from the equator, late-stage TPW driven by low-latitude, late-stage volcanism would be 6-33 degrees, similar to that inferred from the location of paleopolar deposits. 4.4+/-1.3x10^19 kg of young erupted lava can account for the offset of the Dorsa Argentea Formation from the present-day south rotation pole. This mass is consistent with prior mapping-based estimates and would imply a mass release of CO2 by volcanic degassing similar to that in the atmosphere at the present time. The South Polar Layered Deposits are offset from the spin axis in the opposite sense to the other paleopolar deposits. This can be explained by an additional contribution from a plume beneath Elysium. We conclude with a list of observational tests of the TPW hypothesis., Comment: Accepted by Earth and Planetary Science Letters. 3 tables, 8 figures

Data from the Gravity Recovery and Interior Laboratory (GRAIL) mission have revealed that approximately 98 percent of the power of the gravity signal of the Moon at high spherical harmonic degrees correlates with the topography. The remaining 2 percent of the signal, which cannot be explained by topography, contains information about density variations within the crust. These high-degree Bouguer gravity anomalies are likely caused by small-scale (10's of km) shallow density variations. Here we use gravity inversions to model the small-scale three-dimensional variations in the density of the lunar crust. Inversion results from three non-descript areas yield shallow density variations in the range of 100-200 kg/m3. Three end-member scenarios of variations in porosity, intrusions into the crust, and variations in bulk crustal composition were tested as possible sources of the density variations. We find that the density anomalies can be caused entirely by changes in porosity. Characteristics of density anomalies in the South Pole-Aitken basin also support porosity as a primary source of these variations. Mafic intrusions into the crust could explain many, but not all of the anomalies. Additionally, variations in crustal composition revealed by spectral data could only explain a small fraction of the density anomalies. Nevertheless, all three sources of density variations likely contribute. Collectively, results from this study of GRAIL gravity data, combined with other studies of remote sensing data and lunar samples, show that the lunar crust exhibits variations in density by plus or minus 10 percent over scales ranging from centimeters to 100’s of kilometers.

This report is part of a series from the Department of Energy on the use of solar energy in heating buildings. Described here is a new system for year around collection and storage of solar energy. This system has been operated at the University of Virginia for over a year. Composed of an underground hot water storage system and solar collection, the system was tested. Thermal performance results are presented. Analog and digital computer models were used to study the effect of various design modifications on collection and storage performance. (Author/MR)

HPCMP CREATE-AV Conservative Field Finite Element (COFFE) is a modular, extensible, robust numerical solver for the Navier-Stokes equations that invokes modularity and extensibility from its first principles. COFFE implores a flexible, class-based hierarchy that provides a modular approach consisting of discretization, physics, parallelization, and linear algebra components. These components are developed with modern software engineering principles to ensure ease of uptake from a user's or developer's perspective. The Streamwise Upwind/Petrov-Galerkin (SU/PG) method is utilized to discretize the compressible Reynolds-Averaged Navier-Stokes (RANS) equations tightly coupled with a variety of turbulence models. The mathematics and the philosophy of the methodology that makes up COFFE are presented.

This work is a first step in constraining olivine alteration rates on Mars. By bounding dissolution rates of olivine, we can make inferences about the temporal extent of aqueous alteration on the surface. Several hypothesized low-temperature surface aqueous alteration processes include valley network formation from release of groundwater during impact or gully formation from snow melt, both of which may occur over a period of several thousand years. Outflow channels may have formed during brief floods lasting less than a year. If large bodies of water were present during a warm, wet period, large standing bodies of water may have been present for 104 to 109 years. High-temperature alteration processes include hydrothermal activity associated with impacts and magmatic features. These features may remain active for 104 to 106 years. Table 2 summarizes speculated duration times of these aqueous processes. For comparison, we have determined minimum olivine residence times (time until complete dissolution) for 1 mm diameter particles to range from much less than one year to approximately 5,000 years.