Interpretable multi-source data fusion through Latent Variable Gaussian Process
Not Available
Ravi, Sandipp Krishnan↗
Engineering topics
Ghosh, Sayan
Publications and source records attributed to Ghosh, Sayan.
MassiveGNN: Efficient Training via Prefetching for Massively Connected Distributed Graphs
Graph Neural Networks (GNN) are indispensable in learning from graph-structured data, yet their rising computational costs, especially on massively connected graphs, pose significant challenges in terms of execution performance. To tackle this, distributed-memory solutions such as partitioning the graph to concurrently train multiple replicas of GNNs are in practice. However, approaches requiring a partitioned graph usually suffer from communication overhead and load imbalance, even under optimal partitioning and communication strategies due to irregularities in the neighborhood minibatch sampling. This paper proposes practical trade-offs for improving the sampling and communication overheads for representation learn- ing on distributed graphs (using popular GraphSAGE architecture) by developing a parameterized prefetch and eviction scheme on top of the state-of-the-art Amazon DistDGL distributed GNN framework, demonstrating about 15–40% improvement in end-to-end training performance on the NERSC Perlmutter supercomputer for various OGB datasets.
Machine Leanring, high performance comptuing, grap↗
Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays
Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.
Distributed computing, machine learning↗
Analyzing inference workloads for spatiotemporal modeling
Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.
97 MATHEMATICS AND COMPUTING↗
Graph Analytics on Jellyfish topology
Because large unstructured datasets is important for many science domains, distributed graph analytics is critical to many scientists. Unfortunately, obtaining scaling and performance for irregular communication is challenging because contemporary network interconnects are primarily designed to maximize bandwidths of fixed-neighborhoods large-message exchanges (e.g., stencils). Although there is no consensus on the “best” network topologies for irregular communication, unstructured graph-based interconnects can be more suitable. We analyze three popular graph workloads – clustering, pattern enumeration, and traversal — on comparable networks (in terms of resources and costs) constructed from Jellyfish Random Regular, Dragonfly and Fat tree topologies, varying the routing algorithms. Using packet-level simulations, we demonstrate up to 60% improvement in communication time with Jellyfish due to diversity of the short paths between arbitrary endpoints, which can reduce overall network stalls and congestion.
Graph Analytics, network topology, interconnect, H↗
Origin of spin-driven ferroelectricity and effect of external pressure on the complex magnetism of the 6 H perovskite Ba 3 HoRu 2 O 9
The compound Ba 3 HoRu 2 O 9 magnetically orders at 50 K (T N1 ), followed by another complex magnetic ordering at 10.2 K (T N2 ). The second magnetic phase transition was characterized by the coexistence of two competing magnetic ground states associated with two different magnetic wave vectors (K 1 = 0.5 0 0 and K 2 = 0.25 0.25 0). The multiferroicity and magnetoelectric coupling were predicted below T N2 in these 4d-based materials. Here, in this work, we have discussed the origin of spin-driven ferroelectricity, which is not known yet, and the nature of magnetoelectric domains. We have investigated the compound through time-of-flight neutron diffraction, synchrotron x-ray diffraction (XRD), ac susceptibility, frequency-dependent complex dielectric spectroscopy, and dc magnetization under external pressure. We have demonstrated that the noncollinear structure involving two different magnetic ions, Ru (4d) and Ho (4f), breaks the spatial inversion symmetry via inverse Dzyaloshinskii-Moriya (DM) interaction through strong 4d-4f magnetic correlation, which shifts the oxygen atoms and results in nonzero polarization. Such an observation of inverse DM interaction from two different magnetic ions which cause ferroelectricity is rarely observed. The stronger spin-orbit coupling of 4d orbital might play a major role in creating DM interaction of noncollinear spins. We have systematically studied the spin and dipolar dynamics, which exhibit intriguing behavior with shorter coherence lengths of second magnetic phase associated with the k 2 wave vector. The results manifest the development of finite-size magnetoelectric domains instead of true long-range ordering, which justifies the experimentally obtained low value of ferroelectric polarization. The lattice parameters and volume show a sharp anomaly at T N2 obtained by analyzing the temperature-dependence XRD, which is consistent with the ferroelectric transition, predicting a noncentrosymmetric space group, $P\bar{6}2c$, for this compound. Furthermore, we have investigated the effect of external pressure on this complex magnetism. The result reveals an enhancement of ordering temperature by the application of external pressure (~1.6 K/GPa). The external pressure might favor stabilizing the magnetic ground state associated with second magnetic phase. Our study shows an unconventional mechanism of spin-driven ferroelectricity involving inverse DM interaction between Ru (4d) and Ho (4f) magnetic ions due to strong 4d-4f cross coupling.
75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
Analysis and Benchmarking of feature reduction for classification under computational constraints
Abstract Machine learning is most often expensive in terms of computational and memory costs due to training with large volumes of data. Current computational limitations of many computing systems motivate us to investigate practical approaches, such as feature selection and reduction, to reduce the time and memory costs while not sacrificing the accuracy of classification algorithms. In this work, we carefully review, analyze, and identify the feature reduction methods that have low costs/overheads in terms of time and memory. Then, we evaluate the identified reduction methods in terms of their impact on the accuracy, precision, time, and memory costs of traditional classification algorithms. Specifically, we focus on the least resource intensive feature reduction methods that are available in Scikit-Learn library. Since our goal is to identify the best performing low-cost reduction methods, we do not consider complex expensive reduction algorithms in this study. In our evaluation, we find that at quadratic-scale feature reduction, the classification algorithms achieve the best trade-off among competitive performance metrics. Results show that the overall training times are reduced 61%, the model sizes are reduced 6×, and accuracy scores increase 25% compared to the baselines on average with quadratic scale reduction.
97 MATHEMATICS AND COMPUTING↗
pnnl/emp-gnn
Efficient Graph Neural Network for Predicting Molecular Properties software can compare the prediction quality with ab initio DFT results reported by the high-performance state-of-theart NWChem quantum chemistry package [1] through Mean Absolute Error (MAE) obtained by the fitting between DFT and model predictions i.e, MPNN [2]. We also demonstrated the performance benefits by grouping large molecules by atom sizes, and executing GNN models on different types of resources. Since the training times depend on the number of atoms, we demonstrate the impact of distributing the workloads on two GPUs with varying capabilities (e.g., NVIDIA A100 vs. GeForce RTX 2080 Ti) to optimize the efficiency. We are at the precipice of broad adoption of GNNs for molecular property prediction tasks; hence, our work is timely by comparing model prediction against classical approaches with the intent of providing initial screening for specific classes of molecules.
Lee, Hyungro↗
Proxy Applications for Converged Workloads: DMC LDRD Initiative
Modern scientific applications are complicated and require coordination of several components. Proxy application driven software-hardware co-design plays a vital role in driving innovation among the developments of applications, software infrastructure and hardware architecture. Proxy applications are self-contained and simplified codes that are intended to model the performance-critical computations within applications. Applications executing on modern High Performance Computing (HPC) systems are susceptible to network congestion, insufficient memory bandwidth within and across compute nodes, and inadvertent loss of performance due to bugs and unoptimized programming models. Modern numerical simulations and machine learning models play a critical role in studying physical phenomenon under myriad uncertainties. Such applications often exhibit irregular computation and memory accesses at specific regions of the application code, which can contribute to various performance bottlenecks at scale. To mitigate such issues and prepare the next generation hardware for a variety of computation and data movement contingencies, a well-known practice is to consider "proxy" applications as representative motifs for various classes of scientific applications. While there is disagreement in the HPC community on the mechanisms of construction of the proxy applications, there is a strong consensus on their positive impact in co-design. Proxy Applications for Converged Workloads (PACER) is about facilitating software-hardware co-design through proxy applications with the goal of improving the performance of converged science workflows on heterogeneous systems.
97 MATHEMATICS AND COMPUTING↗
Optimizing chemistry for designing oxidation resistant FeCrAl alloys
Abstract Traditionally, FeCrAl alloys played an important role in high-temperature applications due to their ability to form a passive Al oxide film at temperatures above ~ 800 °C. Recently, FeCrAl alloys became of interest for the application of accident tolerant nuclear fuel cladding. This study covers work done at GE Research for better understanding the role of Al, Cr, and Mo in oxidation kinetics and thermodynamics. Several models and commercial prototype alloys have been tested in hydrothermal corrosion autoclave loops, at low temperature steam exposure (~ 400 °C), high temperature steam exposure (~ 1000 °C or higher), and high temperature air exposures. The results provide insights on how chromium and aluminum play a significant role in both high temperature and low temperature oxidation of FeCrAl. Additionally, machine learning tools are used to gain further insights on both predicting future optimized chemistries for balancing the properties of hydrothermal corrosion, low and high temperature steam oxidation, and thermal aging (which is exacerbated due to radiation in a nuclear reactor environment). GE plans to use this framework to further optimize the FeCrAl alloy system for use in nuclear reactor environments. Graphical abstract
Roy, Indranil (ORCID:0000000336124323)↗
Understanding oxidation of Fe-Cr-Al alloys through explainable artificial intelligence
Abstract The oxidation resistance of FeCrAl based on alloying composition and oxidizing conditions is predicted using a combinatorial experimental and artificial intelligence approach. A neural network (NN) classification model was trained on the experimental FeCrAl dataset produced at GE Research. Furthermore, using the SHapley Additive exPlanations (SHAP) explainable artificial intelligence (XAI) tool, we explore how the NN can showcase further material insights that are unavailable directly from a black-box model. We report that high Al and Cr content forms protective oxide layer, while Mo in FeCrAl creates thick unprotective oxide scale that is vulnerable to spallation due to thermal expansion. Graphical abstract
Materials Science↗
ECP-ExaGraph/netAlign
We investigate substituting a parallel, half-approximation for maximum weight matching instead of an exact computation. Our experiments show that the resulting difference in solution quality is negligible. We demonstrate almost a 20-fold speedup using 40 threads on an 8 processor Intel Xeon E7-8870 system and now solve real-world problems in 36 seconds instead of 10 minutes
Khan, Arif↗
Sparse Deep Neural Network Inference using different Programming Models
Sparse deep neural networks have gained increasing attention recently in achieving speedups on inference with reduced memory footprints. However, the real-world applications are little shown with specialized optimizations, yet a wide variety of DNN tasks remain dense without exploiting the advantages of sparsity in networks. Recent work presented by MIT/IEEE/Amazon GraphChallenge has demonstrated significant speedups and various techniques. Still, we find that there is limited investigation of the impact of various Python and C\slash C++ based programming models to explore new opportunities in general cases. In this work, we provide performance evaluation through different programming models using CuPy, cuSPARSE, and OpenMP to discuss the advantages and disadvantages of our sparse implementations on single-GPU and multiple GPUs of NVIDIA DGX-A100 40GB/80GB platforms.
machine learning, HPC↗
Improved Distributed-memory Triangle Counting by Exploiting the Graph Structure
Graphs are ubiquitous in modeling complex systems and representing interactions between entities to uncover structural information of the domain. Traditionally, graph analytics workloads are challenging to efficiently scale (both strong and weak cases) on distributed memory due to the irregular memory-access driven nature (with little or no computations) of the methods. The structure of graphs and their relative distribution over the processing elements poses another level of complexity, making it difficult to attain sustainable scalability across platforms. In this paper, we discuss enhancements to TriC, a distributed-memory implementation of graph triangle counting using Message Passing Interface (MPI), which was featured in the 2020 Graph Challenge competition. We have made some incremental enhancements to TriC, primarily adopting a user-defined buffering strategy to overcome the startup problem for large graphs (by fixing the memory for intermediate data), and experimenting with probabilistic data structures such as bloom filter to improve the query response time for assessing edge existence, at the expense of increasing the overall false positive rate. These adjustments have led to a modest improvements in most cases, as compared to the previous version.
Graph Analytics, HPC↗
Towards scaling community detection on distributed-memory heterogeneous systems
Distributed multi-GPU systems pose significant challenges and opportunities for efficient execution of parallel applications. Graph algorithms are generally characterized by irregular memory accesses, low computation to communication ratios, and load balancing problems that are especially hard to address on multi-GPU systems. Graph community detection is an important problem in the emerging domain of graph analytics with numerous applications. In this paper, we present our ongoing work on distributed-memory multi-GPU implementation for graph community detection. Our work parallelizes the widely used (albeit serial) Louvain method on distributed multi-GPU platforms. Supported by an extensive set of experiments on a multi-GPU enabled supercomputer (OLCF Summit) and a single compute node (Nvidia DGX-2®), we demonstrate competitive performance to existing distributed-memory CPU-based implementation, and up to 6.5 better results than Nvidia RAPIDS® CUGRAPH. To the best of our knowledge, this work represents the first effort for community detection on distributed multi-GPU systems. Our approach and related findings can be extended to numerous other iterative graph algorithms on multi-GPU systems.
97 MATHEMATICS AND COMPUTING↗
EXARL-Easily eXtendable Architecture for Reinforcement Learning
A scalable software framework for reinforcement learning environments and agents/policies used for the Design and Control applications
Ramakrishnaiah, Vinay↗
Single-node Partitioned-Memory for Huge Graph Analytics: Cost and Performance Trade-offs
Nonvolatile memory NVDIMMs, available as Intel Optane, are less expensive than DRAM and bring large byte-addressable storage within reach to many applications. Evaluations on graph analytics have shown promising performance only when DRAM is used as a hardware cache (Memory mode). An open question is whether graph applications can exploit Optane and DRAM directly (AppDirect mode) and achieving better-than-DRAM average bandwidth and run times. We evaluate Optane as a volatile pool on two large-scale graph applications with very different computational patterns, Grappolo and Ripples. We show that AppDirect mode can deliver better-than-DRAM performance, by allocating data structures to Optane and DRAM according to their access characteristics, resulting in higher average memory bandwidth and lower average latency. Memory mode provides DRAM-competitive performance with capacity equal to persistent memory. We demonstrate occasional 4x improvement using the latest AppDirect option and frequently observe competitive performance between Optane AppDirect Memory modes and DRAM.
Ghosh, Sayan↗