Abstract:Many real-world optimisation problems such as hyperparameter tuning in machine learning or simulation-based optimisation can be formulated as expensive-to-evaluate black-box functions. A popular approach to tackle such problems is Bayesian optimisation, which builds a response surface model based on the data collected so far, and uses the mean and uncertainty predicted by the model to decide what information to collect next. In this paper, we propose a generalisation of the well-known Knowledge Gradient acquisition function that allows it to handle constraints. We empirically compare the new algorithm with four other state-of-the-art constrained Bayesian optimisation algorithms and demonstrate its superior performance. We also prove theoretical convergence in the infinite budget limit.

What problem does this paper attempt to address?

The paper aims to address the problem of Bayesian optimization with constraints. Specifically, many practical optimization problems, such as hyperparameter tuning in machine learning and simulation-based optimization, can be formulated as expensive and difficult-to-evaluate black-box functions. For such problems, Bayesian optimization is a commonly used method. It constructs a response surface model based on collected data and uses the predicted mean and uncertainty of this model to decide what information to collect next. However, existing Bayesian optimization methods have shortcomings when dealing with constrained problems. Most methods handle constraints by separately building surrogate models for the objective function and the constraints, and then simply multiplying the acquisition function value of the unconstrained problem by the feasibility probability of the solution. This approach does not fully utilize the value of information about the constraints obtained from sampling. To address this, the paper proposes a generalized Knowledge Gradient (KG) acquisition function—constrained Knowledge Gradient (cKG), which comprehensively considers the value of constraint information. The main contributions of the cKG algorithm include: 1. **Development of the cKG acquisition function**: Capable of handling constraints, unlike existing methods, cKG fully considers the value of constraint information. 2. **Demonstration of an efficient computation method for cKG**: Provides an efficient algorithm for computing cKG. 3. **Proof of theoretical convergence**: In a discrete space, when the budget is unlimited, cKG can converge to the optimal solution. 4. **Extensive experimental validation**: On various test problems, whether the objective function or the constraints have noise, cKG shows performance superior to other existing methods. Through these contributions, the paper provides a new and effective tool for solving Bayesian optimization problems with constraints.

Bayesian Optimisation for Constrained Problems

Practical Bayesian Optimization of Machine Learning Algorithms

Bayesian Optimization for Non-Convex Two-Stage Stochastic Optimization Problems

A Tutorial on Bayesian Optimization

Pseudo-Bayesian Optimization

Process Quality Control Through Bayesian Optimization with Adaptive Local Convergence

Bayesian Optimization with High-Dimensional Outputs

Cost-aware Multi-objective Bayesian optimisation

Simulation Based Bayesian Optimization

Adaptive Expansion Bayesian Optimization for Unbounded Global Optimization

Recent Advances in Bayesian Optimization

Bayesian Optimization with Exponential Convergence

A Bayesian optimization approach for data‐driven mixed‐integer nonlinear programming problems

Expert-guided Bayesian Optimisation for Human-in-the-loop Experimental Design of Known Systems

Investigating Bayesian optimization for expensive-to-evaluate black box functions: Application in fluid dynamics

Enhanced Bayesian Optimization via Preferential Modeling of Abstract Properties

Bayesian Optimization over Discrete and Mixed Spaces via Probabilistic Reparameterization

Human-algorithm collaborative Bayesian optimization for engineering systems

Bayesian Optimization with Shape Constraints

A Data-Driven Bayesian Nonparametric Approach for Black-Box Optimization