Collaborative Pareto Set Learning in Multiple Multi-Objective Optimization Problems

The code is mainly designed to be simple and readable, it contains:

• run_copsl.py is a ~250-line script to run the CoPSL algorithms (including Pareto front visualization);
• run_copsl_gn.py is a ~210-line script to run the CoPSL algorithms with the GradNorm strategy;
• run_psl.py is a ~240-line script to run the PSL algorithms (including Pareto front visualization);
• run_emo.py is a ~100-line script to run the EMO algorithms;
• problem.py includes all the test problems utilized in the paper for running PSL algorithms;
• problem_emo.py also includes all the test problems utilized in the paper for running EMO algorithms;
• model.py contains both single-task and multi-task architecture models;
• utils.py contains several reusable utility functions;
• The folder pf_re contains the files related to the approximate Pareto fronts.

Algorithms

• PSL-LS
• PSL-COSMOS
• PSL-TCH
• PSL-MTCH
• NSGA-II
• NSGA-III
• MOEA/D

Benchmarks

• Two-dimensional synthetic problems: F1 to F6;
• Three-dimensional real-world engineering design problems: RE31, RE32, RE33, RE34, RE37.

Parameters

The parameters specified in the ./run_copsl.py file are as follows:

Parameter	Description
ins_list	List of test problems.
n_run	Number of independent run.
loss_func	The loss function. Options: ls, cosmos, tch, mtch.
n_steps	Number of learning steps.
n_pref_update	Number of sampled preferences per step.
lr	Learning rate.
gamma	The $\gamma$ parameter for cosmos.
device	The device to run the program.
init_seed	Random seed.

Citation

If you find our work helpful in your research, please cite it as:

@inproceedings{shang2024collaborative,
  title={Collaborative Pareto set learning in multiple multi-objective optimization problems},
  author={Shang, Chikai and Ye, Rongguang and Jiang, Jiaqi and Gu, Fangqing},
  booktitle={2024 International Joint Conference on Neural Networks (IJCNN)},
  pages={1--8},
  year={2024},
  organization={IEEE}
}