TLDR: OptunaHub is a new community platform that centralizes black-box optimization (BBO) methods and benchmarks. It provides unified Python APIs (OptunaHub Module), a package registry for community contributions (OptunaHub Registry), and a web interface for discovery (OptunaHub Web). Inspired by Hugging Face Hub, OptunaHub aims to overcome fragmentation in BBO research, foster collaboration, and accelerate advancements by making diverse optimization algorithms and benchmarks easily accessible and reusable.
Black-box optimization (BBO) is a powerful technique driving advancements in various fields, from automating machine learning processes (AutoML) to discovering new materials (Materials Informatics). However, research efforts in BBO have often been scattered across different domains, making it challenging for researchers to share and reuse methods and benchmarks effectively.
To address this fragmentation, researchers Yoshihiko Ozaki, Shuhei Watanabe, and Toshihiko Yanase have introduced OptunaHub, a new community platform designed to centralize BBO methods and benchmarks. Inspired by the success of platforms like Hugging Face Hub, which transformed the machine learning community by providing a central place for models and datasets, OptunaHub aims to create a similar ecosystem for black-box optimization.
OptunaHub offers a unified approach to BBO research through three core components: OptunaHub Module, OptunaHub Registry, and OptunaHub Web. These components work together to promote searchability, reusability, and cross-domain collaboration in BBO.
OptunaHub Module: Unified APIs for Easy Integration
The OptunaHub Module is a Python library that provides easy-to-use, unified APIs. It allows users to effortlessly load registered packages, such as samplers (BBO algorithms) and benchmarks, directly into their Python projects. A key feature is the load_module function, which can import modules like “samplers/autosampler” or “benchmarks/bbob”. This module is designed to be fully compatible with Optuna, a widely used BBO framework known for its robust backend and well-designed APIs. This compatibility means users can easily swap different samplers and benchmark problems, streamlining the process of testing and analyzing optimization results.
OptunaHub Registry: A Hub for Community Contributions
The OptunaHub Registry serves as the gateway for researchers and practitioners to share their own BBO methods and benchmarks with the community. This registry aggregates contributor packages, making them accessible via the Optuna interface. The platform already hosts a diverse collection of samplers, ranging from classical methods like Nelder–Mead to cutting-edge techniques such as HEBO and Bayesian optimization enhanced by LLMs. Many of these methods have been registered by their original authors, including CatCMA, SMAC, SyneTune, and PFNs4BO. As of October 2025, 94 packages have been registered, and monthly downloads have exceeded 100,000, demonstrating the platform’s growing adoption and visibility within the BBO community. Beyond samplers, the registry also accepts benchmarks (like BBOB and HPO/NAS-Bench), pruners, and visualization tools, aiming to collect an ever-increasing array of practical problems.
Also Read:
- ZEROSHOTOPT: A Pretrained Model for Efficient Black-Box Optimization Without Tuning
- AutoMaAS: A Self-Evolving Framework for Multi-Agent AI Systems
OptunaHub Web: Discovering Packages with Ease
OptunaHub Web is the platform’s web interface, crucial for enhancing the searchability and visibility of registered packages. It features individual package pages, automatically generated from each package’s README.md file, which provide detailed information such as author and license details, package summaries, tags, dependency information, API documentation, and code examples. The web interface also offers full-text and tag search capabilities, allowing users to quickly find functionalities of interest. This centralized catalog not only benefits users by making packages easier to find and reuse but also helps contributors promote their work to a broader audience.
In essence, OptunaHub aims to foster a virtuous cycle: high-quality packages attract more users, which in turn encourages further contributions, ultimately accelerating progress in black-box optimization research and its applications. The source code for OptunaHub is publicly available under the Optuna organization on GitHub. For more in-depth information, you can read the full research paper here: OptunaHub: A Platform for Black-Box Optimization.
