PrismBO: A Data-Centric Benchmarking Platform for Composable Transfer Learning in Bayesian Optimization in Dynamic Environments

PrismBO is an open-source software platform designed to facilitate the design, benchmarking, and application of transfer learning for Bayesian optimization (TLBO) algorithms through a modular, data-centric framework.

Getting Started: Getting Started: The key steps in using PrismBO: installation, algorithm building, using benchmarking problems, visualization, and data management for effective transfer learning optimization.

About Us: COLA laboratory is working in computational/artificial intelligence for black-box optimization and decision-making (especially with multiple conflicting objectives).

News: Our system has been applied in various studies, including protein design, hyperparameter optimization...

Video Demonstration

Watch the following video for a quick overview of PrismBO’s capabilities:

Key Features

🧱

Composite Algorithm Design

Modular framework that allows users to construct TLBO algorithms by assembling standardized components—just like building with LEGO bricks.

📚

Robust Data Management

End-to-end data system for storing, retrieving, and managing heterogeneous optimization data across dynamic and evolving problem landscapes.

🔁

Transfer Learning Augmentation

Seamlessly integrates transfer learning into every stage of the BO process to enhance performance and efficiency across multiple tasks.

📊

Benchmarking Across Domains

Provides configurable benchmark generators for HPO, software configuration, and scientific discovery, complete with statistical evaluation tools.

💡

Conversational Web Interface

Interactive LLM-powered UI built with ReactJS for intuitive optimization workflows, experiment tracking, and analytics—no coding required.

🎯

Data-Centric Optimization Agent

Intelligent agent bridges data, TLBO methods, and LLMs for autonomous, adaptive, and lifelong optimization services.

Contents

• Installation Guide
• Quick Start
  • Web User Interface Mode
  • Command Line Mode
• Algorithmic Objects
  • Algorithmic Components
  • Registering a New Algorithmic Object in PrismBO
  • Supported Algorithms
• Benchmark Problems
  • Registering a New Benchmark Problem
  • Synthetic Problem
  • Hyperparameter Optimization Problem
  • Configurable Software Optimization Problem
  • RNA Inverse Design Problem
  • Protein Inverse Folding Problem
  • Parallelization
• Results Analysis
  • Register a New Results Analysis Method
  • Customization Analysis Pipline
  • List of Performance Evaluation Metrics
  • Statistical Measures
  • Wilcoxon Signed-Rank Test
  • Scott-Knott Test
  • A12 Effect Size
  • Critical Difference (CD)
• Data Management
  • Data Storage and Management
  • Similarity Search and Data Reuse
  • Flexible Data Querying
  • Integration with Optimization Tasks
  • Design Considerations
• Visualization
  • Data Filtering and Statistical Visualization
• Command Line
  • Running the Command-Line Interface
• Architecture Overview
  • System Architecture
  • Workflow
  • Optimizer Architecture
  • Data Management
  • Conclusion
• API Reference
• FAQ
  • How do I submit the error information to the maintainer?
  • How do I report a bug?

Contact

Peili Mao

University of Electronic Science and Technology of China

Department of Computer Science

E-mail:

peili.z.mao@gmail.com

Cite

If you have utilized our framework for research purposes, we kindly invite you to cite our publication as follows:

BibTex:

@ARTICLE{PrismBO,
  title = {{PrismBO}: Transfer Optimization System for Bayesian Optimization Using Transfer Learning},
  author = {Author Name and Collaborator Name},
  url = {https://github.com/maopl/PrismBO},
  year = {2024}
}