An R package for tile-based linear algebra built on MPCR, with precision control for individual matrix tiles. It supports parallel CPU execution and optional CUDA GPU execution for operations such as Cholesky factorization and matrix multiplication.
A task-based programming model for parallel execution of R scripts. It uses the COMPSs runtime to manage task dependencies and scheduling across computing resources.
Paper: RCOMPSs: A Scalable Runtime System for R Code Execution on Manycore Systems (preprint).
GPU-accelerated Vecchia approximations for scalable Gaussian-process modeling. It parallelizes conditional computations and provides an R interface for statistical workflows.
Paper: GPU-Accelerated Vecchia Approximations of Gaussian Processes for Geospatial Data using Batched Matrix Computations (ISC 2024; open-access version).
An R package for computations using multiple and mixed numerical precisions. It provides tools for exploring the trade-offs between numerical accuracy, memory use, and computational performance.
GitHub repository | CRAN package
Paper: MPCR: Multi-Precision Computations Package in R (preprint).
The current C++ implementation of ExaGeoStat for high-performance computational geostatistics, with an R interface. It supports spatial modeling workflows, including parameter estimation and prediction, on modern computing architectures.
ExaGeoStat is the older C implementation, and ExaGeoStatR is its R interface. Both legacy repositories have been archived; active development and R package support are now maintained in ExaGeoStatCPP.
Foundational paper: ExaGeoStat: A High Performance Unified Software for Geostatistics on Manycore Systems (IEEE TPDS, 2018).
A toolkit for machine learning at extreme scale. It supports research in parallel and distributed machine-learning computation.
Related MaTEx paper: User-transparent Distributed TensorFlow (preprint).