Scalable Global Optimization via Local Bayesian Optimization

Abstract

Bayesian optimization has recently emerged as a popular method for the sample-efficient optimization of expensive black-box functions. However, the application to high-dimensional problems with several thousand observations remains challenging, and on difficult problems Bayesian optimization is often not competitive with other paradigms. In this paper we take the view that this is due to the implicit homogeneity of the global probabilistic models and an overemphasized exploration that results from global acquisition. This motivates the design of a local probabilistic approach for global optimization of large-scale high-dimensional problems. We propose the $\text{TuRBO}$ algorithm that fits a collection of local models and performs a principled global allocation of samples across these models via an implicit bandit approach. A comprehensive evaluation demonstrates that $\text{TuRBO}$ outperforms state-of-the-art methods from machine learning and operations research on problems spanning reinforcement learning, robotics, and the natural sciences.

Appears in NeurIPS 2019 as a spotlight paper

Cited by in corpus (42)

• Bayesian Optimization is Superior to Random Search for Machine Learning Hyperparameter Tuning: Analysis of the Black-Box Optimization Challenge 2020
• SMAC3: A Versatile Bayesian Optimization Package for Hyperparameter Optimization
• ENTMOOT: A Framework for Optimization over Ensemble Tree Models
• High-Dimensional Bayesian Optimization with Sparse Axis-Aligned Subspaces
• Re-Examining Linear Embeddings for High-Dimensional Bayesian Optimization
• Revisiting Bayesian Optimization in the light of the COCO benchmark
• Learning Search Space Partition for Black-box Optimization using Monte Carlo Tree Search
• A comparison of optimisation algorithms for high-dimensional particle and astrophysics applications
• Black-Box Optimization with Local Generative Surrogates
• Machine Learning with bond information for local structure optimizations in surface science
• Scalable Constrained Bayesian Optimization
• Bag of Baselines for Multi-objective Joint Neural Architecture Search and Hyperparameter Optimization
• Local policy search with Bayesian optimization
• Think Global and Act Local: Bayesian Optimisation over High-Dimensional Categorical and Mixed Search Spaces
• Multi-Objective Bayesian Optimization over High-Dimensional Search Spaces
• Benchmarking the Performance of Bayesian Optimization across Multiple Experimental Materials Science Domains
• High-Dimensional Bayesian Optimisation with Variational Autoencoders and Deep Metric Learning
• Fast Matrix Square Roots with Applications to Gaussian Processes and Bayesian Optimization
• Global Stochastic Optimization of Stellarator Coil Configurations
• Scaling Gaussian Processes with Derivative Information Using Variational Inference
• On the Numerical Performance of Derivative-Free Optimization Methods Based on Finite-Difference Approximations
• AdaDGS: An adaptive black-box optimization method with a nonlocal directional Gaussian smoothing gradient
• Solving Black-Box Optimization Challenge via Learning Search Space Partition for Local Bayesian Optimization
• Cautious Bayesian Optimization for Efficient and Scalable Policy Search
• Facilitating Database Tuning with Hyper-Parameter Optimization: A Comprehensive Experimental Evaluation
• Bayesian Algorithm Execution: Estimating Computable Properties of Black-box Functions Using Mutual Information
• Thermodynamic Consistent Neural Networks for Learning Material Interfacial Mechanics
• Unsupervised Reservoir Computing for Solving Ordinary Differential Equations
• Scalable Thompson Sampling using Sparse Gaussian Process Models
• A portfolio approach to massively parallel Bayesian optimization
• A Scalable Gradient-Free Method for Bayesian Experimental Design with Implicit Models
• GPU Accelerated Exhaustive Search for Optimal Ensemble of Black-Box Optimization Algorithms
• Adaptive Local Bayesian Optimization Over Multiple Discrete Variables
• Bayesian Optimisation for Constrained Problems
• Learning Space Partitions for Path Planning
• Scalable Acceleration for Classification-Based Derivative-Free Optimization
• Online hyperparameter optimization by real-time recurrent learning
• Practical Bayesian Optimization of Objectives with Conditioning Variables
• ChaCha for Online AutoML
• Searching in the Forest for Local Bayesian Optimization
• Learning How to Optimize Black-Box Functions With Extreme Limits on the Number of Function Evaluations
• Optimizing High-Dimensional Physics Simulations via Composite Bayesian Optimization