Preface
Organization
Contents
Foundational Methods Enabling Science in an Integrated Ecosystem
Computational Workflow for Accelerated Molecular Design Using Quantum Chemical Simulations and Deep Learning Models
1 Introduction
2 Computational Workflow for Molecular Design
2.1 Inverse Design of Molecules with Small HOMO-LUMO Gap
3 The DFTB Method
4 Surrogate Models: Graph Convolutional Neural Networks
5 Molecule Generation: Masked Language Model
6 Application: Minimizing the HOMO-LUMO Gap
7 Conclusions and Future Work
References
Self-learning Data Foundation for Scientific AI
1 Introduction
2 Self-learning Data Foundation for AI
3 The Common Metadata Framework (CMF)
3.1 CMF Foundational Pillars
3.2 CMF Architecture
3.3 Distributed CMF with Git-Like Experience
3.4 Integration with AI Frameworks and Experiment Tracking Tools
4 Southbound Intelligence
4.1 Integration with CMF
5 Northbound Intelligence
6 Summary
References
Preconditioners for Batched Iterative Linear Solvers on GPUs
1 Introduction
2 Related Work on Batched Linear Solvers
3 Design and Implementation of Batched Iterative Solvers
4 Design and Implementation of Preconditioners for Batched Iterative Solvers
5 Experimental Evaluation
6 Conclusion
References
Mobility Aware Computation Offloading Model for Edge Computing
1 Introduction
2 Related Work
3 Mobility Aware Computation Offloading Model
3.1 The Proposed Model
3.2 Mobility Prediction
4 Experimentation and Result Analysis
4.1 Simulation Environment
4.2 Experiment Result and Analysis
5 Conclusion
References
Science and Engineering Applications Requiring and Motivating an Integrated Ecosystem
Machine Learning for First Principles Calculations of Material Properties for Ferromagnetic Materials
1 Introduction
2 Surrogate Models
2.1 Linear Model
2.2 Graph Convolutional Neural Network
3 Ferromagnetic Materials
3.1 Solid Solution Binary Alloy Dataset
4 Numerical Section
4.1 Numerical Experiments Using Linear Mixing Model
4.2 Numerical Experiments Using HydraGNN
5 Conclusions and Future Developments
References
A Vision for Coupling Operation of US Fusion Facilities with HPC Systems and the Implications for Workflows and Data Management
1 Introduction
2 Data Needs for Experimental Databases
2.1 Equilibrium Reconstructions via Machine Learning
2.2 Event Labeling via Semi-supervised Learning
2.3 Disruption Prediction via Random Forest Algorithm
2.4 Validation of the TGLF Turbulence Transport Model
2.5 Uncertainty Quantification
3 A Sustainable Community Fusion Simulation Database
4 Vision
5 Conclusion
References
At-the-Edge Data Processing for Low Latency High Throughput Machine Learning Algorithms
1 Introduction
2 Algorithms
2.1 Convolution Method
2.2 Discrete Cosine and Sine Transform Method (DCSTM)
2.3 Streaming Discrete Cosine Transform Method (SDCTM)
2.4 Zero Crossing
3 Results
3.1 Convolution Method
3.2 DCSTM
3.3 Convolution and DCSTM Comparisons
4 Conclusions
References
Implementation of a Framework for Deploying AI Inference Engines in FPGAs
1 Introduction
1.1 The SNL Framework
1.2 Why C++ Templates?
1.3 Why Dynamic Loading of Weights and Biases?
1.4 Why Streaming?
1.5 Overview of SNL Usage
1.6 SNL Limitations, Both Correctable and Intrinsic
2 SNL for Convolution Networks
2.1 Data Widening
2.2 Controlling the Resource and Latency
2.3 Scalability
2.4 Activators
2.5 Layer Implementations: Conv2D
2.6 Layer Implementations: AveragePooling
2.7 Layer Implementations: Dense
2.8 Adding New Layers and Activators
3 Implementation Results: BES Network
4 SNL for Reservoir Networks
5 Conclusion
References
Systems and Software Advances Enabling an Integrated Science and Engineering Ecosystem
Calvera: A Platform for the Interpretation and Analysis of Neutron Scattering Data
1 Background
2 A Neutrons Data Interpretation Platform
3 Science Uses Cases
3.1 Monte Carlo Universe (MCU)
3.2 Advanced Neutron Data Analysis for Quantum Materials (DCA)
3.3 Automatic Structure Refinement Platform (ASRP)
3.4 Quasielastic Neutron Scattering (QClimax)
4 Architecture and Implementation
5 Challenges
5.1 Remote Job Orchestration
5.2 Remote Data Handling
5.3 Authentication/Authorization
6 Current and Future Work
7 Conclusion
A Platform Assessment Criteria
B Evaluated Workflow Systems
References
Virtual Infrastructure Twins: Software Testing Platforms for Computing-Instrument Ecosystems
1 Introduction
2 Science Ecosystems
2.1 Science Workflow Scenarios
2.2 Network Capabilities
3 VIT Concept and Implementation
4 Implementations and Case Studies
4.1 VSNE: Multi-VM for Software Defined Networking
4.2 VFSIE: VM for Federation Stack
4.3 STEM VIT: VM and Native for Microscopes
5 Extensions and Limitations
5.1 Design-to-Deployment Continuum
5.2 Throughput Profiles
6 Conclusions
References
The INTERSECT Open Federated Architecture for the Laboratory of the Future
1 Introduction
2 Related Work
3 The INTERSECT Open Architecture
3.1 Science Use Case Design Patterns
3.2 System of Systems Architecture
3.3 Microservice Architecture
4 Conclusion
References
Real-Time Edge Processing During Data Acquisition
1 Introduction
2 Common Sensor-Driven Workflows
2.1 Data Rates and Data Burden
3 Applications
3.1 X-Ray Ptychography
3.2 Lattice Light Sheet Microscopy
4 High Performance GPU-Enabled Python
5 Streaming Processing Pipelines
5.1 Supercomputing as a Service (SCaaS)
6 Conclusion
References
Towards a Software Development Framework for Interconnected Science Ecosystems
1 Introduction
2 Motivating Use-Case: Scanning Transmission Electron Microscope
3 Approach
3.1 Designing for Interconnected Science Ecosystems
3.2 Digital Twins
3.3 DevSecOps, Process and Software Excellence
4 INTERSECT-SDK
4.1 Dev: Libraries, Services, and Adapters
4.2 DevSecOps
4.3 Digital Twins
5 Related Work
6 Conclusion
References
Deploying Advanced Technologies for an Integrated Science and Engineering Ecosystem
Adrastea: An Efficient FPGA Design Environment for Heterogeneous Scientific Computing and Machine Learning
1 Introduction
2 Background
2.1 FPGA Design
2.2 Xilinx Vitis Toolchain
2.3 Related Work
3 Adrastea Design Environment
3.1 FPGA Building Environment
3.2 IRIS
3.3 DEFFE
3.4 Experiment Setup via Git
3.5 Complete Adrastea Build Flow
4 Example Applications Leveraging Adrastea
4.1 SNS and Other State-of-the-Art RF Models Design Space Exploration
4.2 FFT
4.3 Effectiveness of Adrastea
5 Conclusion and Future Work
References
Toward an Autonomous Workflow for Single Crystal Neutron Diffraction
1 Introduction
2 Background
2.1 Bragg Peak Detection
2.2 Image Segmentation
2.3 Edge Inference
3 Method
3.1 Overview Architecture
3.2 Transfer Learning
3.3 Model Reduction
3.4 Solution Stack
3.5 Continuous Integration
4 Evaluation
4.1 Experiment Setup
4.2 Results
5 Discussion
6 Conclusion
References
Industrial Experience Deploying Heterogeneous Platforms for Use in Multi-modal Power Systems Design Workflows
1 Background
2 Advances and Outlook
3 Workflow Enabled Model Based Engineering
3.1 Requirements and Analysis
3.2 Conceptual Design of the Workflow Framework
3.3 Workflow Framework Implementation
3.4 Workflow Framework Applied: M Ă N Application Readiness
4 Workflow Framework Applied: Reducing Low Cycle Fatigue
4.1 Requirements and Implementation
4.2 Process Workflow Described
4.3 Training Data Generation
4.4 Surrogate Model Creation and Execution
4.5 Validation and Analysis
5 Conclusion
6 Areas for Further Study
References
Self-describing Digital Assets and Their Applications in an Integrated Science and Engineering Ecosystem
1 Introduction
2 Approach
3 Use Cases
3.1 Use Case 1: Management of Distributed Scientific Computing Environments
3.2 Use Case 2: AI Model Sharing Across Two Science Organizations
3.3 Use Case 2a: AI Model Sharing Across Two Science Organizations Using an Existing Catalog
3.4 Use Case 3: Distributed Data Collection Leveraging 5G Infrastructure
3.5 Use Case 4: Federated Learning
4 Architecture and Implementation
5 Conclusion
References
Simulation Workflows in Minutes, at Scale for Next-Generation HPC
1 Introduction
2 Methodology
2.1 Designing an Autonomous System for Simulation Workflows in Minutes
2.2 Hierarchy of Learning
3 Examples of Research Conducted on Summit
3.1 Program exploration on Summit
3.2 Leadership Computing Scale Research in Reinforcement Learning
4 Conclusions
References
Scientific Data Challenges
Machine Learning Approaches to High Throughput Phenotyping
1 Problem Description
2 Methodology
2.1 Set Up Conda Environment
2.2 Install Leptonica
2.3 Install Tesseract
2.4 Install Pytesseract, OpenCV and PlantCV
3 Data Challenges
3.1 Question One
3.2 Question Two
3.3 Question Three
3.4 Question Four
4 Summary and Future Suggestions
References
SMC 2022 Data Challenge: Summit Spelunkers Solution for Challenge 2
1 Introduction
2 Results
2.1 Challenge Question 1: Analysis and Conversion of Data
2.2 Challenge Question 2: Comparison of Usage Patterns
2.3 Challenge Question 3: Analysis of Inter-node Differences
2.4 Challenge Question 4: Prediction of Future Trends
2.5 Challenge Question 5: Additional Trends and Patterns
3 Conclusions
References
Usage Pattern Analysis for the Summit Login Nodes
1 Introduction
2 Data Processing and Feature Extraction
3 Usage Pattern Analysis
3.1 Computing Pattern
3.2 Storage Utilization and Responsiveness
3.3 Submitted Jobs
4 Summary and Future Work
References
Finding Hidden Patterns in High Resolution Wind Flow Model Simulations
1 Background and Related Work
2 Exploratory Data Analysis and Visualization
2.1 Systematic Bias in the Data
2.2 Data Correlation
3 Dimension Reduction
3.1 Methods
3.2 Results
4 Up-Scaling from a Low-Resolution (LR) to High-Resolution (HR) Grid
4.1 Data and Methodology
5 Conclusion and Outlook
A Appendix
References
Investigating Relationships in Environmental and Community Health: Correlations of Environment, Urban Morphology, and Socio-economic Factors in the Los Angeles Metropolitan Statistical Area
1 Introduction
1.1 Case Study Area
1.2 Data Sources
2 Methodology
3 Results and Impressions
4 Conclusions and Future Research
Appendix
References
Patterns and Predictions: Generative Adversarial Networks for Neighborhood Generation
1 Introduction
2 Challenge Question and Proposed Solution
3 Background
4 Approach
4.1 Data Preparation
4.2 Model Architecture
4.3 Optimization
4.4 Model Training
4.5 Evaluation
5 Results and Discussion
6 Conclusions
7 Contributions
References
Author Index