Graph partitioning has been an important tool to partition the work among several processors to minimize the communication cost and balance the workload. While accelerator-based supercomputers are emerging to be the standard, the use of graph partitioning becomes even more important as applications are rapidly moving to these architectures. However, there is no distributed-memory-parallel, multi-GPU graph partitioner available for applications. We developed a spectral graph partitioner, Sphynx, using the portable, accelerator-friendly stack of the Trilinos framework. In Sphynx, we allow using different preconditioners and exploit their unique advantages. We use Sphynx to systematically evaluate the various algorithmic choices in spectral partitioning with a focus on the GPU performance. We perform those evaluations on two distinct classes of graphs: regular (such as meshes, matrices from finite element methods) and irregular (such as social networks and web graphs), and show that different settings and preconditioners are needed for these graph classes. The experimental results on the Summit supercomputer show that Sphynx is the fastest alternative on irregular graphs in an application-friendly setting and obtains a partitioning quality close to ParMETIS on regular graphs. When compared to nvGRAPH on a single GPU, Sphynx is faster and obtains better balance and better quality partitions. Sphynx provides a good and robust partitioning method across a wide range of graphs for applications looking for a GPU-based partitioner.
This report summarizes the activities performed as part of the Science and Engineering of Cybersecurity by Uncertainty quantification and Rigorous Experimentation (SECURE) Grand Challenge LDRD project. We provide an overview of the research done in this project, including work on cyber emulation, uncertainty quantification, and optimization. We present examples of integrated analyses performed on two case studies: a network scanning/detection study and a malware command and control study. We highlight the importance of experimental workflows and list references of papers and presentations developed under this project. We outline lessons learned and suggestions for future work.
Garg, Raveesh; Qin, Eric; Martinez, Francisco M.; Guirado, Robert; Jain, Akshay; Abadal, Sergi; Abellan, Jose L.; Acacio, Manuel E.; Alarcon, Eduard; Rajamanickam, Sivasankaran; Krishna, Tushar
Graph Neural Networks (GNNs) have garnered a lot of recent interest because of their success in learning representations from graph-structured data across several critical applications in cloud and HPC. Owing to their unique compute and memory characteristics that come from an interplay between dense and sparse phases of computations, the emergence of reconfigurable dataflow (aka spatial) accelerators offers promise for acceleration by mapping optimized dataflows (i.e., computation order and parallelism) for both phases. The goal of this work is to characterize and understand the design-space of dataflow choices for running GNNs on spatial accelerators in order for the compilers to optimize the dataflow based on the workload. Specifically, we propose a taxonomy to describe all possible choices for mapping the dense and sparse phases of GNNs spatially and temporally over a spatial accelerator, capturing both the intra-phase dataflow and the inter-phase (pipelined) dataflow. Using this taxonomy, we do deep-dives into the cost and benefits of several dataflows and perform case studies on implications of hardware parameters for dataflows and value of flexibility to support pipelined execution.
The propagation of a wave pulse due to low-speed impact on a one-dimensional, heterogeneous bar is studied. Due to the dispersive character of the medium, the pulse attenuates as it propagates. This attenuation is studied over propagation distances that are much longer than the size of the microstructure. A homogenized peridynamic material model can be calibrated to reproduce the attenuation and spreading of the wave. The calibration consists of matching the dispersion curve for the heterogeneous material near the limit of long wavelengths. It is demonstrated that the peridynamic method reproduces the attenuation of wave pulses predicted by an exact microstructural model over large propagation distances.
With the rapid proliferation of additive manufacturing and 3D printing technologies, architected cellular solids including truss-like 3D lattice topologies offer the opportunity to program the effective material response through topological design at the mesoscale. The present report summarizes several of the key findings from a 3-year Laboratory Directed Research and Development Program. The program set out to explore novel lattice topologies that can be designed to control, redirect, or dissipate energy from one or multiple insult environments relevant to Sandia missions, including crush, shock/impact, vibration, thermal, etc. In the first 4 sections, we document four novel lattice topologies stemming from this study: coulombic lattices, multi-morphology lattices, interpenetrating lattices, and pore-modified gyroid cellular solids, each with unique properties that had not been achieved by existing cellular/lattice metamaterials. The fifth section explores how unintentional lattice imperfections stemming from the manufacturing process, primarily sur face roughness in the case of laser powder bed fusion, serve to cause stochastic response but that in some cases such as elastic response the stochastic behavior is homogenized through the adoption of lattices. In the sixth section we explore a novel neural network screening process that allows such stocastic variability to be predicted. In the last three sections, we explore considerations of computational design of lattices. Specifically, in section 7 using a novel generative optimization scheme to design novel pareto-optimal lattices for multi-objective environments. In section 8, we use computational design to optimize a metallic lattice structure to absorb impact energy for a 1000 ft/s impact. And in section 9, we develop a modified micromorphic continuum model to solve wave propagation problems in lattices efficiently.
We use a nascent data-driven causal discovery method to find and compare causal relationships in observed data and climate model output. We consider ten different features in the Arctic climate collected from public databases on observational and Energy Exascale Earth System Model (E3SM) data. In identifying and analyzing the resulting causal networks, we make meaningful comparisons between observed and climate model interdependencies. This work demonstrates our ability to apply the PCMCI causal discovery algorithm to Arctic climate data, that there are noticeable similarities between observed and simulated Arctic climate dynamics, and that further work is needed to identify specific areas for improvement to better align models with natural observations.
Nonlocal models use integral operators that embed length-scales in their definition. However, the integrands in these operators are difficult to define from the data that are typically available for a given physical system, such as laboratory mechanical property tests. In contrast, molecular dynamics (MD) does not require these integrands, but it suffers from computational limitations in the length and time scales it can address. To combine the strengths of both methods and to obtain a coarse-grained, homogenized continuum model that efficiently and accurately captures materials' behavior, we propose a learning framework to extract, from MD data, an optimal nonlocal model as a surrogate for MD displacements. Our framework guarantees that the resulting model is mathematically well-posed, physically consistent, and that it generalizes well to settings that are different from the ones used during training. The efficacy of this approach is demonstrated with several numerical tests for single layer graphene both in the case of perfect crystal and in the presence of thermal noise.
This work focuses on estimation of unknown states and parameters in a discrete-time, stochastic, SEIR model using reported case counts and mortality data. An SEIR model is based on classifying individuals with respect to their status in regards to the progression of the disease, where S is the number individuals who remain susceptible to the disease, E is the number of individuals who have been exposed to the disease but not yet infectious, I is the number of individuals who are currently infectious, and R is the number of recovered individuals. For convenience, we include in our notation the number of infections or transmissions, T, that represents the number of individuals transitioning from compartment S to compartment E over a particular interval. Similarly, we use C to represent the number of reported cases.
A new copper equation of state is developed utilizing the available experimental data in addition to recent theoretical calculations. Semi-empirical models are fit to the data and the results are tabulated in the SNL SESAME format. Comparison to other copper EOS tables are given, along with recommendations of which tables provide the best accuracy.
Constructing accurate statistical models of critical system responses typically requires an enormous amount of data from physical experiments or numerical simulations. Unfortunately, data generation is often expensive and time consuming. To streamline the data generation process, optimal experimental design determines the 'best' allocation of experiments with respect to a criterion that measures the ability to estimate some important aspect of an assumed statistical model. While optimal design has a vast literature, few researchers have developed design paradigms targeting tail statistics, such as quantiles. In this project, we tailored and extended traditional design paradigms to target distribution tails. Our approach included (i) the development of new optimality criteria to shape the distribution of prediction variances, (ii) the development of novel risk-adapted surrogate models that provably overestimate certain statistics including the probability of exceeding a threshold, and (iii) the asymptotic analysis of regression approaches that target tail statistics such as superquantile regression. To accompany our theoretical contributions, we released implementations of our methods for surrogate modeling and design of experiments in two complementary open source software packages, the ROL/OED Toolkit and PyApprox.
This report details the results of a three-fold investigation of sensitivity analysis (SA) for machine learning (ML) explainability (MLE): (1) the mathematical assessment of the fidelity of an explanation with respect to a learned ML model, (2) quantifying the trustworthiness of a prediction, and (3) the impact of MLE on the efficiency of end-users through multiple users studies. We focused on the cybersecurity domain as the data is inherently non-intuitive. As ML is being using in an increasing number of domains, including domains where being wrong can elicit high consequences, MLE has been proposed as a means of generating trust in a learned ML models by end users. However, little analysis has been performed to determine if the explanations accurately represent the target model and they themselves should be trusted beyond subjective inspection. Current state-of-the-art MLE techniques only provide a list of important features based on heuristic measures and/or make certain assumptions about the data and the model which are not representative of the real-world data and models. Further, most are designed without considering the usefulness by an end-user in a broader context. To address these issues, we present a notion of explanation fidelity based on Shapley values from cooperative game theory. We find that all of the investigated MLE explainability methods produce explanations that are incongruent with the ML model that is being explained. This is because they make critical assumptions about feature independence and linear feature interactions for computational reasons. We also find that in deployed, explanations are rarely used due to a variety of reason including that there are several other tools which are trusted more than the explanations and there is little incentive to use the explanations. In the cases when the explanations are used, we found that there is the danger that explanations persuade the end users to wrongly accept false positives and false negatives. However, ML model developers and maintainers find the explanations more useful to help ensure that the ML model does not have obvious biases. In light of these findings, we suggest a number of future directions including developing MLE methods that directly model non-linear model interactions and including design principles that take into account the usefulness of explanations to the end user. We also augment explanations with a set of trustworthiness measures that measure geometric aspects of the data to determine if the model output should be trusted.
The typical topology optimization workflow uses a design domain that does not change during the optimization process. Consequently, features of the design domain, such as the location of loads and constraints, must be determined in advance and are not optimizable. A method is proposed herein that allows the design domain to be optimized along with the topology. This approach uses topology and shape derivatives to guide nested optimizers to the optimal topology and design domain. The details of the method are discussed, and examples are provided that demonstrate the utility of this approach.
Ship tracks are quasi-linear cloud patterns produced from the interaction of ship emissions with low boundary layer clouds. They are visible throughout the diurnal cycle in satellite images from space-borne assets like the Advanced Baseline Imagers (ABI) aboard the National Oceanic and Atmospheric Administration Geostationary Operational Environmental Satellites (GOES-R). However, complex atmospheric dynamics often make it difficult to identify and characterize the formation and evolution of tracks. Ship tracks have the potential to increase a cloud's albedo and reduce the impact of global warming. Thus, it is important to study these patterns to better understand the complex atmospheric interactions between aerosols and clouds to improve our climate models, and examine the efficacy of climate interventions, such as marine cloud brightening. Over the course of this 3-year project, we have developed novel data-driven techniques that advance our ability to assess the effects of ship emissions on marine environments and the risks of future marine cloud brightening efforts. The three main innovative technical contributions we will document here are a method to track aerosol injections using optical flow, a stochastic simulation model for track formations and an automated detection algorithm for efficient identification of ship tracks in large datasets.