LibMOP_Pareto_Front_Learning is an open-source library built on PyTorch for Learning Pareto Front of Multi-Objective Problem (MOP).
- [Jun 04 2024]: Added support for Hyper-Transformer (ArXiv 2024). Many thanks to the author's help @Tuan.
- [Jun 04 2024]: Added support for STCH (ArXiv 2024). Many thanks to the author's help @Tuan.
- [Jun 04 2024]: Added support for TCH,LOG,PROD,UTILITY,COSINE (Neural Network 2024). Many thanks to the author's help @Tuan.
- [Jun 03 2024]: Added support for EPO (ICML 2020). Many thanks to the author's help @Mahapatra.
- [Jun 02 2024]: Added support for LS (ICLR 2021). Many thanks to the author's help @AvivNavon.
LibMOP_Pareto_Front_Learning currently supports the following algorithms with MLP architecture (params = 61102) and Transformer architecture (params = 61202), hidden_dim = 100. We calculate Hypervolume Difference (HVD) to each of algorithms:
| Optimization Strategies | Venues | CVX2 | Arguments |
|---|---|---|---|
| Linear Scalarization (LS) | ICLR 2021 | 0.00052 | --model_type mlp --solver LS |
| Exact Pareto Optimal Search (EPO) | ICML 2020 | 0.00060 | --model_type mlp --solver EPO |
| Weighted Chebyshev (TCH) | Neural Network 2024 | 0.00095 | --model_type mlp --solver TCH |
| Log Scalarization (LOG) | Neural Network 2024 | 0.00097 | --model_type mlp --solver LOG |
| Product Scalarization (PROD) | Neural Network 2024 | 0.00090 | --model_type mlp --solver PROD |
| Utility Scalarization (UTILITY) | Neural Network 2024 | 0.00046 | --model_type mlp --solver UTILITY |
| Cosine Scalarization (COSINE) | Neural Network 2024 | 0.00161 | --model_type mlp --solver COSINE |
| Smooth Weighted Chebyshev (STCH) | ArXiv 2024 | 0.00047 | --model_type mlp --solver STCH |
Transformer architecture (Hyper-Trans)
| Optimization Strategies | Venues | CVX2 | Arguments |
|---|---|---|---|
| Linear Scalarization (LS) | ICLR 2021 | 0.00057 | --model_type trans --solver LS |
| Exact Pareto Optimal Search (EPO) | ICML 2020 | 0.00054 | --model_type trans --solver EPO |
| Weighted Chebyshev (TCH) | Neural Network 2024 | 0.00052 | --model_type trans --solver TCH |
| Log Scalarization (LOG) | Neural Network 2024 | 0.00078 | --model_type trans --solver LOG |
| Product Scalarization (PROD) | Neural Network 2024 | 0.00080 | --model_type trans --solver PROD |
| Utility Scalarization (UTILITY) | Neural Network 2024 | 0.00045 | --model_type trans --solver UTILITY |
| Cosine Scalarization (COSINE) | Neural Network 2024 | 0.00731 | --model_type trans --solver COSINE |
| Smooth Weighted Chebyshev (STCH) | ArXiv 2024 | 0.00039 | --model_type trans --solver STCH |
If you have any question or suggestion, please feel free to contact us by raising an issue or sending an email to trananhtuan23012000@gmail.com.
LibMOP_Pareto_Front_Learning is released under the MIT license.