Preference-Aware Bayesian Optimization for Interactive Decision Making.

Area of Science:

• Engineering
• Computer Science
• Operations Research

Background:

• Real-world optimization problems often involve multiple conflicting objectives.
• Approximating the Pareto front is a common strategy for multiobjective optimization (MOO).
• Existing methods struggle to incorporate real-time decision-maker preferences, limiting practical application.

Purpose of the Study:

• To propose a novel preference-aware Bayesian optimization (PABO) framework.
• To enable interactive decision-making by integrating decision-maker feedback.
• To efficiently find a single, most preferred solution from the Pareto-optimal set.

Main Methods:

• Developed a PABO framework that embeds preference information into candidate solution generation.
• Dynamically balances exploration of uncertain regions and exploitation of preference-aligned solutions.
• Incorporates real-time decision-maker feedback throughout the optimization process.

Main Results:

• PABO achieves comparable or superior solution quality compared to state-of-the-art methods.
• Significantly fewer expensive evaluations are required with PABO.
• Demonstrates improved optimization efficiency and reduced costs.

Conclusions:

• PABO offers a more feasible technical approach for practical multiobjective optimization problems.
• The framework effectively addresses the challenge of incorporating real-time preferences.
• PABO enhances efficiency and reduces costs in complex optimization scenarios.