Fup0

Fup0 is a lightweight software library that provides a modular framework for performing Fast Uniform Probability (FUP) sampling in high‑dimensional numerical optimization tasks. The library was originally developed by the Computational Methods Group at the Institute of Advanced Systems in 2024. It implements a single‑pass polling algorithm that eliminates the need for explicit probability mass functions, thereby reducing computational overhead in large‑scale simulations.

The core design of Fup0 centers around a series of interfaces that allow users to plug in arbitrary objective functions, constraints, and adaptive step‑size schemes. A built‑in caching mechanism stores previously evaluated points, which can be reused when the sampling domain is refined. The library is written in C++ and exposes bindings for Python, R, and MATLAB, making it suitable for both research and industry applications.

Fup0 is distributed under the open‑source MIT license. Its accompanying documentation includes a quick‑start tutorial, an API reference, and a collection of example scripts that demonstrate its use in Bayesian inference, evolutionary algorithms, and uncertainty quantification. The library’s dependency list is intentionally minimal: it relies only on the standard C++17 library and the Eigen linear algebra package.

Subsequent releases of the project have focused on expanding its capabilities. Fup1 introduced support for parallel execution on multi‑core architectures via OpenMP, while Fup2 added GPU acceleration through CUDA. Current community contributions include a plugin interface for domain‑specific samplers and a profiler that helps identify bottlenecks in the sampling workflow.