Amazon SageMaker Automatic Model Tuning: Scalable Gradient-Free Optimization

Abstract

Tuning complex machine learning systems is challenging. Machine learning models typically expose a set of hyperparameters, be it regularization, architecture, or optimization parameters, whose careful tuning is critical to achieve good performance. To democratize access to such systems, it is essential to automate this tuning process. This paper presents Amazon SageMaker Automatic Model Tuning (AMT), a fully managed system for black-box optimization at scale. AMT finds the best version of a machine learning model by repeatedly training it with different hyperparameter configurations. It leverages either random search or Bayesian optimization to choose the hyperparameter values resulting in the best-performing model, as measured by the metric chosen by the user. AMT can be used with built-in algorithms, custom algorithms, and Amazon SageMaker pre-built containers for machine learning frameworks. We discuss the core functionality, system architecture and our design principles. We also describe some more advanced features provided by AMT, such as automated early stopping and warm-starting, demonstrating their benefits in experiments.

Machine Learning for Electronically Excited States of Molecules

“…This procedure is beneficial if it is repeated several times while narrowing the space of hyperparameters every time. As hyperparameter optimization is an optimization problem, algorithms, such as Bayesian optimization − have been designed for this task and are frequently applied for deep ML models and kernel methods. Because of the tedious procedure of manually tuning hyperparameters or the expert knowledge, which is required in most cases to find optimal hyperparameters, ML models have been designed which automatically learn optimal hyperparameters and only require little human intervention, see, e.g., refs ,, , , − .…”

IoT data analytics in dynamic environments: From an automated machine learning perspective

“…Amazon SageMaker [150] is an AutoML tool built on Amazon Web Services (AWS). It involves automated model tuning as a major module.…”

Automated Reinforcement Learning (AutoRL): A Survey and Open Problems

“…The next set of approaches involved are those which inherently have a notion of sequential decision making. Bayesian Optimization (BO, Mockus (1974), Jones et al (1998), Brochu et al (2010)) is one of the most popular approaches to date, used for industrial applications (Golovin et al, 2017;Balandat et al, 2020;Perrone et al, 2021) and a variety of scientific experimentation (Frazier & Wang, 2015;Hernández-Lobato et al, 2017;Li et al, 2018;Griffiths & Hernández-Lobato, 2020;Tran et al, 2021;van Bueren et al, 2021). For RL applications, one of the most prominent uses of BO was for tuning AlphaGo's hyperparameters, which include its core Monte Carlo Tree Search (MCTS) (Browne et al, 2012) hyperparameters and time-control settings.…”