RAPIDS Highlights

The RAPIDS SciDAC Institute for Computer Science and Data has objective of assisting Office of Science (SC) application teams in overcoming computer science and data challenges in the use of DOE supercomputing resources to achieve science breakthroughs. 

The following slides contain highlights of recent scientific achievements:

• An AI-enabled Framework for Earthquake Early Warning by Forecasting Ground Motions

• Generative Modeling of Regular and Irregular Time Series Data via Koopman VAEs

• Does In-Context Operator Learning Generalize to Domain-Shifted Settings?

• SuperBench: A Super-Resolution Benchmark Dataset for Scientific Machine Learning

• Robustifying State-space Models for Long Sequences via Approximate Diagonalization

• Efficient Value Distribution Estimation from INR

• MDLoader: A Hybrid Model-driven Data Loader

• Extrapolation of Nuclear Observables in the Limit

• Uncertainty Quantification-enabled inversion of nuclear responses

• Semi-Implicit Neural ODEs for Learning Chaotic Systems

• Parametric Sensitivities of Oceans Using Neural Surrogates

• EFIT-PRIME: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D 

• Physics-informed neural surrogate with Green’s function augmentation for consistent plasma equilibrium reconstruction

• Guaranteeing Error Bounds in Neural Network Based Autoencoders

• Machine Learning for the Time Evolution of Aerosol Microphysics

• Continuous Physical Field Reconstruction from Sparse Observations

• The Art of Sparsity: Mastering High-Dimensional Tensor Storage

• TensorSearch: High dimensional and Parallel Tensor Search Engine

• Sustainable Staging Transport (SST)

• HDF5 Sustainable Strategies for Longevity

• Efficient Streaming Analysis of Plasma Data

• Scaling Millions of Quantum Chemical Calculations

• Streaming Data in HPC Workflows Using ADIOS

• Portable Staging for GPU-Based In-Situ Workflows

• Optimizing the Processing and Storage of Radio Astronomy Data

• Variable Order Multigrid Data Compression With MGARD

• I/O in WRF: Evaluating Modern Parallel I/O Techniques

• AIIO: Using Artificial Intelligence for Job-Level and Automatic I/O Performance Bottleneck Diagnosis

• A Machine Learning Approach for Differential Data Reduction

• Interactive Visualization for Atmosphere Modeling

• FunMC2: Visualization of Contours in Uncertain Fields

• zfp: Compressed Floating-Point Arrays

• A Comparative Study of the Perceptual Sensitivity of Topological Visualizations to Feature Variations

• In Situ Particle Tracing for MPAS-Ocean

• High order remapping from MPAS ocean to ROMS

• Topological Optimization with Big Steps

• Adaptively Placed Multi-grid Scene Representation Network

• Explorable INR for Ensemble Simulation

• Regularized Ensemble Scene Representation Network

• Uncertainty-Aware Flow-Based Surrogate Model

• Energy Efficient Image Segmentation through Reinforcement Learning

• Loosely Coupled Asynchronous In Situ

• Flash-X Performance Orchestration

• Accelerating Large-Scale Atomistic td-DFTB Simulations Using GPU Offloading

• MFDn Communication Optimization

• sKokkos: Enabling Kokkos with Transparent Device Selection on Heterogeneous Systems using OpenACC

• CCAMP: Integrated Translation and Optimization for OpenACC and OpenMP

• IRIS Reimagined: Advancements in Intelligent Runtime System for Task-Based Programmin

• CHARM-SYCL: New Unified Programming Environment for Multiple Kinds of Accelerators

• MAPredict: Static Analysis Driven Memory Access Prediction Framework for Modern CPUs and GPUs

• Clacc: OpenACC Support for C/C++

• Performance portability solutions for GPUs and CPUs to track reconstruction kernels

• Verification of OpenMP SIMD using CIVL

• HPC Code Migration Using Large Language Models

• ChatBLAS: A Portable AI-Generated BLAS Library

• ytopt: A ML-based Autotuning for Energy Efficiency

• Scaling physics-informed hard constraints with mixture-of-experts

• Stability-Aware Training of Neural Network Interatomic Potentials with Differentiable Boltzmann Estimators

• Graph Methods for Lattice QCD Calculations

• Neural Spectral Methods: Self-supervised learning in the spectral domain

• Machine Learned Magnetohdrodynamic Reduced-Order Models

• Lagrangian Particle Tracing for Next Generation Architectures

• Verification of Structured Code Refactoring

• Parallel Data Aggregation for Neutrino Event Analysis

• Plasma-Material Coupling with ADIOS

• Surrogate Modeling for Spatio-Temporal Data from Earth System Models

• Accelerating Density Functional Theory Computations

• Accelerating Sparse Triangular Solves in Fusion Science Codes through One-Sided Communication

• Accelerating HEP Data Analysis on HPC Platforms

• Accelerating I/O for VPIC Simulations

• Galaxy-Scale Strong Gravitational Lens Detection

• Integrating Human Perception with Computational Power for Guided Exploratory Data Analysis

• Data Management and Visualization for Plasma Physics

• Automated Parallel Event Generation and Analysis

• Accelerating Event Reconstruction for Liquid Argon TPC Neutrino Detectors

• Enabling Global Adjoint Tomography at scale through next-generation I/O 

• Graphical Model Structure Learning at Unprecedented Scale 

• Global Particle-in-Cell Simulation of Fusion Plasmas 

• Multivariate, Temporal Visual Analytics for Climate Model Analysis

• Accelerating Earthquake Detection 

• Accelerating Weather Research Forecasting Simulations with Deep Neural Network Surrogates 

• Accelerating Computational Kernels of Tokamak Simulations (with FASTMath) 

• Performance Optimization for Multiscale Gyrokinetic Turbulence 

• Accelerating HEP Event Generation and Analysis on HPC Systems

• Accelerating Fusion Fission Simulations

• In situ Viz Unlocks Unsteady Dynamics at Extreme Scale

• Roofline-Based Modeling in Intel Advisor 

• Improving Network Throughput with Global Communication Reordering 

• Improving Collective Reduction Performance On Manycore Architectures 

• Deep Stack Program Optimization 

• Autonomic Data Movement for Data Staging-based In-Situ Workflows

• Robust IO Performance Modeling in Leadership-Class Systems by Automated Change Detection

• In Situ Compression Artifact Removal in Scientific Data Using Deep Transfer Learning

• Error-controlled Reduction and Retrieval for Scientific Data

• FTK: A Spacetime Meshing Framework for Robust and Scalable Feature Tracking

• Analysis of GPU Data Access Patterns on Complex Geometries for D3Q19 Lattice Boltzmann Algorithm

• In Situ Compression Artifact Removal in Scientific Data Using Deep Transfer Learning and Experience Replay

• Machine learning-based application grouping and knowledge extraction for HPC I/O

• Fast neural Poincaré maps for toroidal magnetic fields

• Cosmic Inference: Constraining Parameters With Observations and Highly Limited Number of Simulations

• IExchange: Asynchronous Communication and Termination Detection for Iterative Algorithms

• Asynchronous and Load-Balanced Union-Find for Distributed and Parallel Scientific Data Visualization and Analysis

• RISE: Reducing I/O Contention in Staging-based Extreme-Scale In-situ Workflows

• Stabilized Neural Ordinary Differential Equations for Long-Time Forecasting of Dynamical Systems

• Porting and Tuning HBPS Applications with TAU and APEX

• MAPredict: Static Analysis Driven Memory Access Prediction Framework for Modern CPUs and GPUs

• Analyzing the Impact of Lossy Data Reduction on Volume Rendering of Cosmology Data

• VDL-Surrogate: A View-Dependent Visualization Surrogate for Ensemble Simulations

• Near-Term Climate Prediction and Explanation

• Leveraging One-Sided Communication for GPU-accelerated Sparse Triangular Solvers

• FES-ASCR Partnership for Research on Superfacility Workflows

• Chemical reaction networks and opportunities for machine learning

• Learning differentiable solvers for systems with hard constraints

• An ecosystem for digital reticular chemistry

• Error Estimation for Random Fourier Features

• Steady-state collisional radiative closures using random forest regressors

• Understanding HPC I/O Performance Models

• NN-EFIT: Physics-constrained Plasma equilibrium reconstruction for magnetically confined Fusion

• Machine Learning Based Volumetric Compression

• Deep Bayesian Climate Emulator

• Adaptive Optimization of Model Training for Multiple GPUs

• How to Achieve High I/O Throughput on Perlmutter

• Understanding Data Access Patterns for Distributed Science Collaboration

• Intelligent Draining to Reduce I/O Contention in Staging-based Workflows

• Adaptive Elasticity for In Situ Analysis and Visualization

• Region-adaptive, Error-controlled Scientific Data Compression for Scientific Data

• Error-bounded compression with preservation of derived quantities

• Mitigate Data Management Challenges for the Exa.TrkX Workflow

• A Scalable Distributed Workflow for Quantum Chemical Prediction of UV/Vis Absorption Spectra for Organic Molecules

• Extreme-Scale Ensembles of Iterative Random Forests

• Scalable Training of Graph Convolutional Neural Networks using the ADIOS Scientific Data Management Library

• HPC Storage Service Autotuning Using Variational-Autoencoder-Guided Asynchronous Bayesian Optimization

• Using Multi-Resolution Data to Accelerate Neural Network Training Time

• Fast Merge Tree Computation via SYCL

• Mesh repartitioning and flattening added to VisIt

• VDL-Surrogate: A View-Dependent Latent-based Model for Parameter Space Exploration of Ensemble Simulations

• GNN-Surrogate: A Hierarchical and Adaptive Graph Neural Network for Parameter Space Exploration of Unstructured-Mesh Ocean Simulations

• Neural Stream Functions

• Deep Hierarchical Super-Resolution for Scientific Data Reduction and Visualization

• IDLat: An Importance-Driven Latent Generation Method for Scientific Data

• Latent Representations for Particle Feature Exploration

• LANL used RAPIDS2 Production Visualization in Producing Science Video

• Improved Parallel Rendering Performance with GPU-Based Image Compression

• VTK-m Demonstrates Performance Portability

• Analyzing the Impact of Lossy Data Reduction on Volume Rendering of Cosmology Data

• Automating Optimization of Generator Tuning Parameters

• Machine Learning-Driven Adaptive OpenMP For Portable Performance on Heterogeneous Systems

• Porting and Tuning HBPS Applications with TAU and APEX

• Performance Portability of Sparse Computational Methods on GPU-Accelerated Architectures

• ML-based Performance Portability for Density Functional Theory in HPC Environments

• Optimized Preprocessing For Scientific Deep Learning Applications

• Multi-GPU Parallel Sparse Triangular Solver 

• Leveraging One-Sided Communication for Sparse Triangular Solvers on traditional CPUs

• Critical Path Model for Sparse Triangular Solvers

• FES-ASCR Partnership for Research on Superfacility Workflows

• EFIT-AI: Performance Portable GPU-accelerated Reconstruction

• CCAMP: Integrated Translation and Optimization for OpenACC and OpenMP

• IRIS: A Portable Task Runtime System for Extremely Heterogeneous Computing

• Porting HEP Event Reconstruction to GPU-based Heterogeneous Systems

• MAPredict: Static Analysis Driven Memory Access Prediction Framework for Modern CPUs and GPUs

• Analysis of GPU Data Access Patterns on Complex Geometries for D3Q19 Lattice Boltzmann Algorithm

• Verifying Fortran Programs with CIVL

• Verification of Data Race Freedom for OpenMP Programs