RoMA: Robust Model Adaptation for Offline Model-based Optimization

Implementation of RoMA: Robust Model Adaptation for Offline Model-based Optimization (NeurIPS 2021).

Setup

conda create -n roma python=3.7
conda activate roma
pip install -r requirement.txt

Run experiments

python run.py --task [TASK]

where available tasks are TASKS=[ant, superconductor, dkitty, hopper, gfp, molecule].

Citation

@inproceedings{
    yu2021roma,
    title={RoMA: Robust Model Adaptation for Offline Model-based Optimization},
    author={Yu, Sihyun and Ahn, Sungsoo and Song, Le and Shin, Jinwoo},
    booktitle={Advances in Neural Information Processing Systems},
    year={2021},
}

References

• Design-bench
• Design-baselines
• Adversarial weight perturbation