Bayesian Optimization with Gradients

Abstract

Bayesian optimization has been successful at global optimization of expensive-to-evaluate multimodal objective functions. However, unlike most optimization methods, Bayesian optimization typically does not use derivative information. In this paper we show how Bayesian optimization can exploit derivative information to decrease the number of objective function evaluations required for good performance. In particular, we develop a novel Bayesian optimization algorithm, the derivative-enabled knowledge-gradient (dKG), for which we show one-step Bayes-optimality, asymptotic consistency, and greater one-step value of information than is possible in the derivative-free setting. Our procedure accommodates noisy and incomplete derivative information, comes in both sequential and batch forms, and can optionally reduce the computational cost of inference through automatically selected retention of a single directional derivative. We also compute the d-KG acquisition function and its gradient using a novel fast discretization-free technique. We show d-KG provides state-of-the-art performance compared to a wide range of optimization procedures with and without gradients, on benchmarks including logistic regression, deep learning, kernel learning, and k-nearest neighbors.

arXiv commentAdvances in Neural Information Processing Systems 30 (NIPS), 2017

Cited by in corpus (34)

• A Tutorial on Bayesian Optimization
• pySOT and POAP: An event-driven asynchronous framework for surrogate optimization
• Scaling Gaussian Process Regression with Derivatives
• Bayesian optimization with improved scalability and derivative information for efficient design of nanophotonic structures
• Time series forecasting with Gaussian Processes needs priors
• Local policy search with Bayesian optimization
• ITSO: A novel Inverse Transform Sampling-based Optimization algorithm for stochastic search
• Scaling Gaussian Processes with Derivative Information Using Variational Inference
• Applying Bayesian Optimization with Gaussian Process Regression to Computational Fluid Dynamics Problems
• ProBO: Versatile Bayesian Optimization Using Any Probabilistic Programming Language
• A Batched Scalable Multi-Objective Bayesian Optimization Algorithm
• Information Directed Sampling for Linear Partial Monitoring
• Semi-supervised Embedding Learning for High-dimensional Bayesian Optimization
• Excursion Search for Constrained Bayesian Optimization under a Limited Budget of Failures
• Linearly Constrained Gaussian Processes with Boundary Conditions
• Learning Effective Loss Functions Efficiently
• Knowledge Gradient for Selection with Covariates: Consistency and Computation
• Learning Personalized Thermal Preferences via Bayesian Active Learning with Unimodality Constraints
• Correcting boundary over-exploration deficiencies in Bayesian optimization with virtual derivative sign observations
• BISTRO: Berkeley Integrated System for Transportation Optimization
• Using Distance Correlation for Efficient Bayesian Optimization
• SLEIPNIR: Deterministic and Provably Accurate Feature Expansion for Gaussian Process Regression with Derivatives
• Neural Generative Models for Global Optimization with Gradients
• Uncertainty Quantification for Bayesian Optimization
• Constrained, Global Optimization of Functions with Lipschitz Continuous Gradients
• Deep kernel learning for integral measurements
• Computing the Value of Computation for Planning
• Graduated Optimization of Black-Box Functions
• Using Known Information to Accelerate HyperParameters Optimization Based on SMBO
• Active learning for enumerating local minima based on Gaussian process derivatives
• A tree-based radial basis function method for noisy parallel surrogate optimization
• Accelerating Experimental Design by Incorporating Experimenter Hunches
• Automatic Calibration of Dynamic and Heterogeneous Parameters in Agent-based Model
• A New Bayesian Optimization Algorithm for Complex High-Dimensional Disease Epidemic Systems