Approximation-Aware Bayesian Optimization

2024-06-06

Neural Information Processing Systems

AI Summary

High-dimensional Bayesian optimization (e.g., molecular design) suffers from expensive function evaluations and degraded acquisition quality due to standard sparse variational Gaussian processes (SVGP), which prioritize global posterior fidelity over decision-relevant accuracy. This paper proposes a utility-calibrated variational inference framework that, for the first time, unifies Gaussian process approximation and decision-oriented data acquisition into a single joint optimization problem—subject to computational budget constraints—to guarantee acquisition-policy optimality. The framework is compatible with theoretically grounded acquisition functions—including expected improvement (EI) and knowledge gradient—and supports trust-region (e.g., TuRBO) and batch-optimization settings. On high-dimensional benchmark tasks in control and molecular design, it significantly outperforms standard SVGP: achieving superior solutions with fewer function evaluations, faster convergence, and higher overall optimization efficiency.

Technology Category

• Reasoning under Uncertainty: Stochastic Optimization
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

High-dimensional Bayesian optimization (BO) tasks such as molecular design often require 10,000 function evaluations before obtaining meaningful results. While methods like sparse variational Gaussian processes (SVGPs) reduce computational requirements in these settings, the underlying approximations result in suboptimal data acquisitions that slow the progress of optimization. In this paper we modify SVGPs to better align with the goals of BO: targeting informed data acquisition rather than global posterior fidelity. Using the framework of utility-calibrated variational inference, we unify GP approximation and data acquisition into a joint optimization problem, thereby ensuring optimal decisions under a limited computational budget. Our approach can be used with any decision-theoretic acquisition function and is compatible with trust region methods like TuRBO. We derive efficient joint objectives for the expected improvement and knowledge gradient acquisition functions in both the standard and batch BO settings. Our approach outperforms standard SVGPs on high-dimensional benchmark tasks in control and molecular design.

Problem

• Reducing computational cost in high-dimensional Bayesian optimization
• Improving data acquisition efficiency in sparse variational Gaussian processes
• Unifying GP approximation and data acquisition for optimal decisions

Innovation

• Modifies SVGPs for better data acquisition
• Unifies GP approximation and data acquisition
• Compatible with trust region methods