In plain words: Google's Vizier service tunes settings for research and production systems by guessing how changes affect results and trying the most promising ones next. Its default tuning algorithm stayed reliable across many kinds of real problems, holding up well against well-known industry tools on standard tests.
Abstract
Google Vizier has performed millions of optimizations and accelerated numerous research and production systems at Google, demonstrating the success of Bayesian optimization as a large-scale service. Over multiple years, its algorithm has been improved considerably, through the collective experiences of numerous research efforts and user feedback. In this technical report, we discuss the implementation details and design choices of the current default algorithm provided by Open Source Vizier. Our experiments on standardized benchmarks reveal its robustness and versatility against well-established industry baselines on multiple practical modes.
Xingyou Song, Qiuyi Zhang, Chansoo Lee, Emily Fertig, Tzu-Kuo Huang, Lior Belenki, Greg Kochanski, Setareh Ariafar, Srinivas Vasudevan, Sagi Perel, Daniel Golovin
arXiv:2408.11527 · cs.LG, cs.AI, math.OC · submitted Aug 21, 2024 · updated Dec 6, 2024
abstract · pdf · html · Google DeepMind Technical Report. Code can be found in https://github.com/google/vizier
