Bandit Limited Discrepancy Search and Application to Machine Learning Pipeline Optimization
Abstract
Optimizing a machine learning (ML) pipeline has been an important topic of AI and ML. Despite recent progress, pipeline optimization remains a challenging problem, due to potentially many combinations to consider as well as slow training and validation. We present the BLDS algorithm for optimized algorithm selection (ML operations) in a fixed ML pipeline structure. BLDS performs multi-fidelity optimization for selecting ML algorithms trained with smaller computational overhead, while controlling its pipeline search based on multi-armed bandit and limited discrepancy search. Our experiments on well-known classification benchmarks show that BLDS is superior to competing algorithms. We also combine BLDS with hyperparameter optimization, empirically showing the advantage of BLDS.
Cite
Text
Kishimoto et al. "Bandit Limited Discrepancy Search and Application to Machine Learning Pipeline Optimization." AAAI Conference on Artificial Intelligence, 2022. doi:10.1609/AAAI.V36I9.21263Markdown
[Kishimoto et al. "Bandit Limited Discrepancy Search and Application to Machine Learning Pipeline Optimization." AAAI Conference on Artificial Intelligence, 2022.](https://mlanthology.org/aaai/2022/kishimoto2022aaai-bandit/) doi:10.1609/AAAI.V36I9.21263BibTeX
@inproceedings{kishimoto2022aaai-bandit,
title = {{Bandit Limited Discrepancy Search and Application to Machine Learning Pipeline Optimization}},
author = {Kishimoto, Akihiro and Bouneffouf, Djallel and Marinescu, Radu and Ram, Parikshit and Rawat, Ambrish and Wistuba, Martin and Palmes, Paulito P. and Botea, Adi},
booktitle = {AAAI Conference on Artificial Intelligence},
year = {2022},
pages = {10228-10237},
doi = {10.1609/AAAI.V36I9.21263},
url = {https://mlanthology.org/aaai/2022/kishimoto2022aaai-bandit/}
}