Official implementation of our paper "LiBOG: Lifelong Learning for Black-Box Optimizer Generation", which is accepted at IJCAI 2025. In this work, we tackle the problem of lifelong learning from a sequence of BBO problem distributions to generate high-performance BBO optimizer using a novel LiBOG, which achieves good forward transferring, mild catastrophic forgetting and stable learning.
LiBOG: Lifelong Learning for Black-Box Optimizer Generation, Jiyuan Pei, Yi Mei, Jialin Liu and Mengjie Zhang. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI) 2025. pp. 8912--8920.
@inproceedings{LiBOG,
title = {LiBOG: Lifelong Learning for Black-Box Optimizer Generation},
author = {Pei, Jiyuan and Mei, Yi and Liu, Jialin and Zhang, Mengjie},
booktitle = {Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, {IJCAI-25}},
publisher = {International Joint Conferences on Artificial Intelligence Organization},
editor = {James Kwok},
pages = {8912--8920},
year = {2025},
month = {8},
note = {Main Track},
doi = {10.24963/ijcai.2025/991},
url = {https://doi.org/10.24963/ijcai.2025/991},
}
├── run_lifelong.py # Entry point for training/testing
├── options.py # Running options for training/testing
├── execute/ # Function code for training/testing
├── model/ # Model definitions
├── expr/ # Code for expression
├── dataset/ # Dataset loading and preprocessing of BBO problems
├── env/ # RL environment of BBO optimizer for solving problems
├── utils/ # Utility functions and helpers
├── population/ # About solution population
├── pbo_env/ # Classic BBO optimizers
├── requirements.txt # Python dependencies
└── README.md # This file
Built upon: Symbol – used and extended with MIT License
The dependencies of this project are listed in requirements.txt. You can install them using the following command.
pip install -r requirements.txt
python run_lifelong.py --train --ll_training_method LiBOG --run_name test_code