Summary¶
The mfpml package is a toolkit for multi-fidelity probabilistic machine learning, Bayesian optimization based on single-/multi-fidelity data resources. This package is design to provide a platform of replicating existing single-/multi-fidelity Bayesian methods based on Gaussian process regression; and also a handy tool for developing new methods in the field of multi-fidelity probabilistic machine learning. Overall, this package has two main features:
Basic methods: fundamental implementations of popular methods for example Gaussian process regression, Co-Kriging and corresponding extensions
Advanced Methods: advanced methods for Bayesian optimization, for instance, riable-fidelity optimization, active learning reliability analysis, multi-fidelity reliability analysis.
Basic Methods¶
The mfpml package provides essential tools and implementations across four key areas:
Design of Experiments
Provides efficient strategies for designing experiments to reduce cost and improve performance.
Functions
A collection of commonly used functions for experimental analysis and performance evaluation.
Models
Kriging Model: A Gaussian process-based surrogate model for single-fidelity data.
mfpml.models.gaussian_process
Multi-Fidelity Kriging Model: Extends Kriging to incorporate multi-fidelity data for better predictions.
mfpml.models.mf_gaussian_process
Optimization
Evolutionary Algorithms: A metaheuristic approach for solving complex optimization problems.
mfpml.optimization.evolutionary_algorithms
Single-Fidelity Bayesian Optimization (SFBO): Bayesian optimization for expensive single-fidelity black-box functions.
mfpml.optimization.sfbo
Multi-Fidelity Bayesian Optimization (MFBO): An advanced Bayesian optimization framework that leverages multi-fidelity data.
mfpml.optimization.mfbo
Advanced methods¶
The mfpml package also includes advanced methods, implemented based on the following publications. These methods push the boundaries of multi-fidelity modeling, optimization, and reliability analysis.
Multi-Fidelity Lower Confidence Bounding Jiang, Ping, et al. “Variable-fidelity lower confidence bounding approach for engineering optimization problems with expensive simulations.” AIAA Journal, 57(12), 2019, pp. 5416-5430.
Focus: Multi-fidelity optimization for computationally expensive problems.
Enhanced Multi-Fidelity Optimization Cheng, Ji, Qiao Lin, and Jiaxiang Yi. “An enhanced variable-fidelity optimization approach for constrained optimization problems and its parallelization.” Structural and Multidisciplinary Optimization, 65(7), 2022, p. 188.
Focus: Improved constrained optimization and parallel implementation.
Adaptive Kriging-Based Reliability Analysis Yi, Jiaxiang, et al. “Efficient adaptive Kriging-based reliability analysis combining new learning function and error-based stopping criterion.” Structural and Multidisciplinary Optimization, 62, 2020, pp. 2517-2536.
Focus: Enhanced Kriging reliability analysis with adaptive learning.
Active-Learning Multi-Fidelity Kriging for Reliability Analysis Yi, Jiaxiang, et al. “An active-learning method based on multi-fidelity Kriging model for structural reliability analysis.” Structural and Multidisciplinary Optimization, 63, 2021, pp. 173-195.
Focus: Combining active learning with multi-fidelity Kriging for reliability analysis.
Fidelity Selection Strategy for Multi-Fidelity Reliability Analysis Yi, Jiaxiang, Yuansheng Cheng, and Jun Liu. “A novel fidelity selection strategy-guided multifidelity Kriging algorithm for structural reliability analysis.” Reliability Engineering & System Safety, 219, 2022, p. 108247.
Focus: Novel fidelity selection strategies to enhance reliability analysis.
Community Support¶
If you encounter any issues, bugs, or problems with this package, please:
Raise an issue on the official GitHub page
Contact the authors via email
Your feedback is highly valued and will help improve the package!
License¶
Copyright © 2024, Jiaxiang Yi and Ji Cheng All rights reserved.
This project is licensed under the MIT License. For details, refer to the LICENSE file included with the repository.