Driving Scientific and Engineering Discoveries Through the Integration of Experiment, Big Data, and Modeling and Simulation (Communications in Computer and Information Science)

Preface
Organization
Contents
Computational Applications: Converged HPC and Artificial Intelligence
Randomized Multilevel Monte Carlo for Embarrassingly Parallel Inference
	1 Introduction
		1.1 The Sweet and the Bitter of Bayes
	2 Technical Details of the Methodology
		2.1 Multilevel Monte Carlo
		2.2 Randomized Multilevel Monte Carlo
		2.3 Multi-index Monte Carlo
	3 Motivating Example
		3.1 Example of Problem
		3.2 Numerical Results
	4 Conclusion and Path Forward
	References
Maintaining Trust in Reduction: Preserving the Accuracy of Quantities of Interest for Lossy Compression
	1 Challenges in Lossy Compression for Physics Simulations
	2 Background of Error-Controlled Lossy Compression
	3 Error-Controlled Lossy Compression in High-Dimensional Space
	4 Error-Controlled Lossy Compression on Nonuniform Grids
	5 Error-Controlled Lossy Compression for QoIs
	6 Future Work
	7 Conclusion
	References
Applying Recent Machine Learning Approaches to Accelerate the Algebraic Multigrid Method for Fluid Simulations
	1 Introduction
	2 Background
		2.1 Overview of Algebraic Multigrid
		2.2 A Dataset of Sparse Systems from 3D Unstructured Meshes
	3 Overview of the Recently Proposed Methods
		3.1 Deep Residual Feed-Forward Network for 2D Structured Grid Problems (Greenfeld et al. ch3Greenfeld2019)
		3.2 Graph Neural Networks for Unstructured Problems (Luz et al. ch3Luz2020)
	4 Results
		4.1 For the Model in Greenfeld et al.
		4.2 For the Model in Luz et al.
	5 Conclusion
	A  Dataset
	References
Building an Integrated Ecosystem of Computational and Observational Facilities to Accelerate Scientific Discovery
	1 Introduction
	2 Background and Context: Connecting Science Facilities
	3 Overview of the ORNL Federated Science Edge Ecosystem
	4 Resources
		4.1 Instrument Interfaces
		4.2 Compute
		4.3 Storage
		4.4 Network
	5 Federation Services and Policies
		5.1 Resource Management
		5.2 Command and Control
		5.3 Workflows
		5.4 Dashboard
		5.5 Data Movement
		5.6 Data Management
		5.7 Identity Management
		5.8 Policy and Governance
	6 Use Case: Autonomous Microscopy at CNMS
		6.1 Scanning Probe Microscopy
		6.2 Electron Microscopy
	7 Conclusions
	References
Advanced Computing Applications: Use Cases that Combine Multiple Aspects of Data and Modeling
Fast and Accurate Predictions of Total Energy for Solid Solution Alloys with Graph Convolutional Neural Networks
	1 Introduction
	2 Physical System - Solid Solution Binary Alloys
	3 Graph Convolutional Neural Networks (GCNNs)
		3.1 Software Implementation
	4 Dataset Description
	5 Use of Federated Instruments, Compute, and Storage
	6 Numerical Results
		6.1 Comparison Between Computational Times for First Principles Calculations and DL Models
		6.2 Comparison Between Statistical Models for Predictive Performance
	7 Conclusions
	References
Transitioning from File-Based HPC Workflows to Streaming Data Pipelines with openPMD and ADIOS2
	1 The Need for Loosely-Coupled Data Pipelines
		1.1 The IO Bottleneck – A Challenge for Large-Scale IO
		1.2 From Monolithic Frameworks to Loosely-Coupled Pipelines
		1.3 Related Work
	2 Building a System for Streaming IO
		2.1 Loosely Coupled Data Processing Pipelines Built via Streaming
		2.2 Impact in an Increasingly Heterogeneous Compute Landscape
		2.3 OpenPMD and ADIOS2: Scientific Self-description and Streaming
	3 Data Distribution Patterns
		3.1 Properties Found in a Performant Distribution Pattern
		3.2 Chunk Distribution Algorithms
	4 Evaluation for Two Streaming Setups
		4.1 Streaming as Basis for an Asynchronous IO Workflow
		4.2 A Staged Simulation-Analysis Pipeline: Setup
		4.3 A Staged Simulation-Analysis Pipeline: Evaluation
	5 Summary and Outlook
	References
Understanding and Leveraging the I/O Patterns of Emerging Machine Learning Analytics
	1 Introduction
	2 Machine Learning Patterns
		2.1 Overview
		2.2 Optimization Opportunities
		2.3 Automation Opportunities
	3 Data Management Vision
		3.1 Model Metadata
		3.2 Add Query Capabilities over the Model Metadata
		3.3 Positioning and Impact
	4 Conclusions
	References
Secure Collaborative Environment for Seamless Sharing of Scientific Knowledge
	1 Introduction
		1.1 Background and Motivation
		1.2 Organization
	2 Use Case Scenarios
		2.1 Data Quality Assessment from Small-Angle Neutron Scattering (SANS) Instrument
		2.2 Data Collection
		2.3 ML Models and Privacy
	3 Machine Learning Model Training
		3.1 Model Setup
		3.2 Model Training
		3.3 Differential Privacy Cost
	4 Secure Inference
		4.1 Secure MPC Setup
		4.2 Secure MPC Performance
		4.3 Secure MPC Cost
	5 Conclusion and Future Work
	References
The Convergence of HPC, AI and Big Data in Rapid-Response to the COVID-19 Pandemic
	1 Background
		1.1 Problem Statement
		1.2 Collaborative Response
		1.3 HPC, AI and Big Data Case Studies
	2 Case Study 1: High Performance Computing
		2.1 Problem Statement
		2.2 Methodology
		2.3 Results
	3 Case Study 2: Artificial Intelligence
		3.1 Problem Statement
		3.2 Methodology
		3.3 Results
	4 Case Study 3: Big Data
		4.1 Problem Statement
		4.2 Methodology
		4.3 Results
	5 Lessons Learned
		5.1 Scientific Gaps and Benefits
		5.2 Technology Gaps and Benefits
	6 Summary
	References
High-Performance Ptychographic Reconstruction with Federated Facilities
	1 Introduction
	2 Background
	3 Ptychography Workflow with Federated Resources
		3.1 Automated Light Source Workflow Execution and Coordination
		3.2 Transparent Remote Function Calls and Data Transfers
		3.3 Accelerated Ptychographic Image Reconstruction
	4 Experimental Results
		4.1 Optimum GPU Configuration
		4.2 End-to-End Workflow Evaluation
	5 Related Work
	6 Conclusion
	References
Machine-Learning Accelerated Studies of Materials with High Performance and Edge Computing
	1 The Scientific Question: Correlated Quantum Materials
	2 The Dynamic Cluster Approximation Quantum Monte Carlo Method, DCA
	3 ML-Accelerated Simulations and Feedback Loop Between Simulations and Experiments
		3.1 Current Practices
		3.2 Proof of Principles
	4 Challenges in Current Workflow
		4.1 An Observation: Two Data Sources and Two Edges
		4.2 Programming Languages Inconsistency for Different Tasks in the Workflow
		4.3 Lack of Standardized ML Model Specification Format
		4.4 Heterogeneity in Hardware and Software Architectures on Different Edge Devices
	5 Opportunities and Needs for Development
		5.1 A Unified Edge Capable of Serving both HPC and Experimental Data Sources
		5.2 Compatibility of HPC and ML Software Stack; Package Management Tools for Both HPC and ML Library Dependencies
		5.3 Workflow Tools and Policies on Edge Computers
		5.4 Standardized ML Model Specification Format
	6 Conclusions and Outlook
	References
Advanced Computing Systems and Software: Connecting Instruments from Edge to Supercomputers
A Hardware Co-design Workflow for Scientific Instruments at the Edge
	1 Introduction
	2 Background
	3 Hardware Programming Ecosystem
		3.1 Chisel Hardware Construction Language
		3.2 Open-Source Hardware Development Ecosystems
	4 Co-design Workflow for Hardware Libraries
	5 Conclusion
	References
NREL Stratus - Enabling Workflows to Fuse Data Streams, Modeling, Simulation, and Machine Learning
	1 Introduction
	2 Advanced Computing at the National Renewable Energy Laboratory
		2.1 ESIF High Performance Computing Data Center
		2.2 NREL Stratus Cloud Computing
	3 Competitive Positioning
		3.1 HPC
		3.2 Cloud Computing
		3.3 Edge Computing
	4 Hybrid Support of Real-Time Data Vision at Scale
		4.1 Overview of the Workflow
		4.2 Workflow Components and Positioning
	5 Challenges to Supporting the Vision
	6 Building Toward the Vision at NREL
	References
Braid-DB: Toward AI-Driven Science with Machine Learning Provenance
	1 Introduction
	2 Background
		2.1 Provenance Needs in AI for Science
		2.2 Developments in Provenance Concepts for Machine Learning
		2.3 Globus Flows
	3 Approach
		3.1 Contributions
		3.2 Provenance Structure
	4 Architecture
		4.1 Software Performance Targets
		4.2 Software Components
		4.3 Software Implementation
	5 Case Studies
		5.1 Provenance Flow Capture for Training DNNs in X-Ray Science
		5.2 Serial Synchrotron X-Ray Crystallography
		5.3 The Mascot Workflow
	6 Performance
	7 Future Work
	8 Conclusion
	References
Lessons Learned on the Interface Between Quantum and Conventional Networking
	1 Introduction
	2 Generic Quantum-Conventional Network Harness
	3 Scientific Use Case
		3.1 Entanglement Distribution
		3.2 Prototype Network Architecture
	4 Deployed Network
		4.1 Time Synchronization
		4.2 Experimental Implementation
		4.3 Bandwidth Allocation
	5 Summary of Lessons Learned
	6 Conclusion
	References
Use It or Lose It: Cheap Compute Everywhere
	1 Introduction
	2 Motivation
		2.1 Premium and Freemium
		2.2 Data Movement
	3 Hardware Design Points
		3.1 FPGA
		3.2 Many-Core
		3.3 Cheap CPUs
		3.4 ASICs and ASIC-Hybrid
	4 Target Market and Use Cases
		4.1 Opportunities Better Suited for Specialization Offload
		4.2 Opportunities for Ancillary Offload
	5 Open Questions
		5.1 Balancing Power, Performance and Cost
		5.2 The Right Level of Abstraction
		5.3 The Memory Problem
		5.4 Tying It All Together
	6 Conclusions
	References
Deploying Advanced Computing Platforms: On the Road to a Converged Ecosystem
Enabling ISO Standard Languages for Complex HPC Workflows
	1 Introduction
		1.1 The Role of Standards in HPC
		1.2 Performance and Programming Models
	2 ISO Standard C++ Parallelism for HPC Workloads
		2.1 ISO C++ Today
		2.2 Evolving ISO C++
	3 ISO Standard Fortran Parallelism for HPC Workloads
		3.1 ISO Fortran Today
		3.2 Evolving ISO Fortran
	4 Conclusion
	References
Towards Standard Kubernetes Scheduling Interfaces for Converged Computing
	1 Introduction
	2 Technology Building Blocks
		2.1 Flux
		2.2 Kube-scheduler
		2.3 Node Feature Discovery
	3 A Design Space for Scheduler Composition
		3.1 Looser Composition Provided by Custom Controllers
		3.2 Tighter Composition Enabled by Scheduling Framework
		3.3 Augmented Tight Composition Demands an API Extension
		3.4 API Semantics Must Accommodate Fundamental Mismatches
	4 KubeFlux Design and Challenges
		4.1 Addressing Mismatches in Workloads and Resource Models
		4.2 Resource Sharing
		4.3 KubeFlux Plug-in Scheduler
	5 Experimental Work
		5.1 Overhead of KubeFlux over Flux
		5.2 Overhead of KubeFlux over Kubernetes
	6 Related Work
		6.1 Cooperative Scheduling
		6.2 Integrating HPC Schedulers in Container-Based Environments
	7 Summary and Future Work
	References
Scaling SQL to the Supercomputer for Interactive Analysis of Simulation Data
	1 Introduction
	2 RAPIDS and the BlazingSQL Software Architecture
	3 Implementation of Communications via UCX in BlazingSQL
		3.1 False Starts for Implementing the UCX API in an Application Code
		3.2 Final Implementation of UCX Communications
	4 Performance Results
		4.1 Performance of UCX vs. IPoIB
		4.2 Multi-node Performance
	5 Implications for Future and Emerging HPC Platforms
	6 Conclusion
	References
NVIDIA\'s Cloud Native Supercomputing
	1 Introduction
	2 Solution: Cloud Native Supercomputing Design Principles
	3 Implementation Principles and Technologies
		3.1 Bluefield Data Processing Unit
		3.2 Multi-tenant Isolation: Toward a Zero-Trust Architecture
		3.3 Services
		3.4 In-Network Computing: Offloading Capabilities
	4 Summary
	References
Scientific Data Challenges
Smoky Mountain Data Challenge 2021: An Open Call to Solve Scientific Data Challenges Using Advanced Data Analytics and Edge Computing
	1 Introduction
	2 Challenge 1: Unraveling Hidden Order and Dynamics in a Heterogeneous Ferroelectric System Using Machine Learning
		2.1 Background
		2.2 Dataset Description
		2.3 Challenges of Interest
	3 Challenge 2: Finding Novel Links in COVID-19 Knowledge Graph
		3.1 Background
		3.2 Introduction
		3.3 Dataset Description
		3.4 Challenges of Interest
	4 Challenge 3: Synthetic-to-Real Domain Adaptation for Autonomous Driving
		4.1 Background
		4.2 Dataset Description
		4.3 Challenges of Interest
	5 Challenge 4: Analyzing Resource Utilization and User Behavior on Titan Supercomputer
		5.1 Background
		5.2 Dataset Description
		5.3 Challenges of Interest
	6 Challenge 5: Sustainable Cities: Socioeconomics, Building Types, and Urban Morphology
		6.1 Background
		6.2 Dataset Description
		6.3 Challenges of Interest
	7 Challenge 6: Where to Go in the Atomic World
		7.1 Background
		7.2 Dataset Description
		7.3 Challenges of Interest
	8 Challenge 7: Increased Image Spatial Resolution for Neutron Radiography
		8.1 Background
		8.2 Dataset Description
		8.3 Challenges of Interest
	9 Challenge 8: High Dimensional Active Learning for Microscopy of Nanoscale Materials
		9.1 Background
		9.2 Dataset Description
		9.3 Challenges of Interest
	10 Conclusions
	References
Advanced Image Reconstruction for MCP Detector in Event Mode
	1 Introduction
	2 Image Reconstruction
		2.1 Event Clustering
		2.2 Incident Neutron Back-Tracing
	3 Results
		3.1 Clustering Results
		3.2 Image Reconstruction
	4 Discussion
		4.1 Event Clustering Analysis
		4.2 Image Reconstruction with Different Models
	5 Summary
	References
An Study on the Resource Utilization and User Behavior on Titan Supercomputer
	1 Introduction
	2 Exploratory Data Analysis
		2.1 Data Preprocessing
		2.2 Data Correlation
		2.3 Data Clustering
	3 Time Series Analysis
		3.1 Seasonality
		3.2 GPU Hardware-Related Issues
		3.3 Predictive Model
	4 Conclusions
	References
Recurrent Multi-task Graph Convolutional Networks for COVID-19 Knowledge Graph Link Prediction
	1 Introduction
	2 Related Work
		2.1 COVID-19 Knowledge Graphs
		2.2 Temporal Link Prediction
	3 Methodology
		3.1 Problem Formation
		3.2 Model
	4 Experiments
		4.1 Datasets
		4.2 Preprocessing
		4.3 Analysis
		4.4 Baseline Models
		4.5 Experimental Setup
		4.6 Results
	5 Conclusions
	References
Reconstructing Piezoelectric Responses over a Lattice: Adaptive Sampling of Low Dimensional Time Series Representations Based on Relative Isolation and Gradient Size
	1 Introduction and Background
	2 Proposed Solution
	3 Approach
	4 Results
	5 Contributions
	References
Finding Novel Links in COVID-19 Knowledge Graph Using Graph Embedding Techniques
	1 Introduction
	2 Methodology
		2.1 Dataset Curation
		2.2 Data Visualization
		2.3 Exploratory Data Analysis
		2.4 Methods for Link Prediction
		2.5 Importance of Predicted Links
	3 Results and Discussions
	4 Conclusion
	References
Exploring the Spatial Relationship Between Demographic Indicators and the Built Environment of a City
	1 Background
		1.1 Introduction
		1.2 Study Area and Data Sources
	2 Methodology
		2.1 Challenge 1: What is the Distribution of Commercial, Industrial, and Residential Buildings Within Each Block Group? Do These Distributions Correlate with Building Age? Building Value? Building Size?
		2.2 Challenge 2: Using Temperature Data from a Source of the Participant’s Choosing, Are There Locations Within the City That Tend to Be Warmer Than Others? How Does This Relate to Building Density and Building Type?
		2.3 Challenge 3: How Does the Built Environment and the Local Scale Experience of Heat Co-vary with Socio-economic and Demographic Characteristics of Residents?
	3 Results
		3.1 Challenge 1: Distribution of Building Types and Their Correlation with Building Characteristics (Age, Value and Size)
		3.2 Challenge 2: Spatial Pattern of Temperature and Its Correlation with Building Type and Density
		3.3 Challenge 3: Correlation Between Socio-economic and Demographic Characteristics and the Built Environment
	4 Conclusion
	Appendix
	References
Atomic Defect Identification with Sparse Sampling and Deep Learning
	1 Introduction
	2 Background
	3 Methods
	4 Results
	5 Conclusion
	References
Author Index