Bandit Limited Discrepancy Search and Application to Machine Learning Pipeline Optimization
Akihiro Kishimoto, Djallel Bouneffouf, Radu Marinescu, Parikshit Ram, Ambrish Rawat, Martin Wistuba, Paulito Palmes, Adi Botea
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.
BibTeX
@inproceedings{aaai2022_banditlimiteddis,
title = {Bandit Limited Discrepancy Search and Application to Machine Learning Pipeline Optimization},
author = {Akihiro Kishimoto and Djallel Bouneffouf and Radu Marinescu and Parikshit Ram and Ambrish Rawat and Martin Wistuba and Paulito Palmes and Adi Botea},
booktitle = {AAAI 2022},
year = {2022}
}