💫We propose CAMO (Convergence-Aware Multi-fidelity Optimization), a novel MFBO framework based on Fidelity Differential Equations (FiDEs). CAMO explicitly captures the convergence behavior of the objective function through rigorous theoretical analysis, enabling more efficient optimization across fidelity levels. Our approach allows for automatic adaptation to varying fidelity levels without extensive manual intervention, making it particularly suitable for practical MFBO scenarios where the highest and lowest fidelity levels may be unknown or difficult to specify.
The performance of CAMO in terms of simple regret on the Borehole, Colville, and Himmelblau datasets is as follows. It can be observed that our method converges to the optimal solution the fastest across all three datasets, highlighting the importance of convergence awareness. More experimental results can be found in the paper.
project_root/
├── README.md
├── requirements.txt
├── Acquisition_function
├── assert (some figure and data)
├── Data_simulation (cost function, real application and synthetic mf function)
├── Experiment
├── FidelityFusion_models(surrogate model)
├── GaussionProcess
We recommend using a virtual environment.
git clone https://github.com/Fillip1233/CAMO.git
cd CAMO
We support simulation data of 10+, Cost_function can be selected from pow_10, linear, or log
cd Experiment\CMF
python CMF_norm.py -- "data_name" --cost_type "pow_10"
Visualization of experimental results
We provide graphic code
python Graph_for6data.py
In the real data experiments, we used Python to call MATLAB scripts. Therefore, when running the real data experiments, please ensure that MATLAB and the MATLAB Engine API for Python are installed on your computer. We provide two usage examples in real-world scenarios: HeatedBlock and VibratePlate
python CMF_real.py -- "data_name" --cost_type "pow_10"
- Reporting bugs. To report a bug, simply open an issue in the GitHub Issues.
- Suggesting enhancements. To submit an enhancement suggestion, including completely new features or minor improvements on existing features, please open an issue in the GitHub Issues.
- Pull requests. If you made improvements to FidelityFusion, fixed a bug, or had a new example, feel free to send us a pull-request.
- Asking questions. To get help on how to use FidelityFusion or its functionalities, you can open a discussion in the GitHub.
💥Please cite our paper if you find it helpful :)