Research map / Probabilistic and scientific modeling / Bayesian optimization
Preference-Aware Constrained Multi-Objective Bayesian Optimization (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes PAC-MOO, a preference-aware constrained multi-objective Bayesian optimization method that weights an information-gain acquisition function (MESMOC-style) over objectives and constraints and adds a feasibility-seeking probability term.
Problem
Constrained multi-objective black-box optimization with user preferences over objectives, large design spaces, and rare feasible inputs that are expensive to find.
Approach
Injects designer preferences into an information-theoretic constrained MOO acquisition function by a convex combination of per-function entropy terms, and adds a product-of-constraint-probabilities acquisition for finding feasible regions.
Key results
- On the modified OSY problem, PAC-MOO with no preferences typically outperforms all baselines in PHV, per the paper's figures.
- PAC-MOO with preferences finds feasible designs with higher values of the preferred objective than all baselines (Figures 1b, 1d, 1f).
- Preference weighting can reduce hypervolume in some cases, while PHV remains competitive with PAC-MOO-0.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use PAC-MOO with equal preferences instead of baseline methods | Constrained multi-objective optimization · SCVR | PHV | Helped |
| use PAC-MOO with equal preferences instead of baseline methods | Constrained multi-objective optimization · HCR | PHV | Helped |
| use PAC-MOO with equal preferences instead of baseline methods | Constrained multi-objective optimization · OSY | PHV | Helped |
| add preferences for the target objective | PAC-MOO · SCVR | maximum discovered feasible value of the preferred objective | Helped |
| add preferences for the target objective | PAC-MOO · HCR | maximum discovered feasible value of the preferred objective | Helped |
| add preferences for the target objective | PAC-MOO · OSY | maximum discovered feasible value of the preferred objective | Helped |
| add preferences for the target objective | PAC-MOO · SCVR and HCR | PHV | Mixed |
| use a version of USEMOC instead of evolutionary baselines | USEMOC · Reported benchmarks | performance | Helped |
Surprising findings
- Emphasizing preferred objectives can lower hypervolume, since PHV measures overall Pareto front quality rather than the preferred region.
Limitations
- Preference-weighted performance can trade off hypervolume against the preferred objective.
- Evaluation is limited to a modified OSY problem in this abstract; full details are deferred to an extended version.
Experimental setup
Datasets: modified OSY problem SCVR HCR
Benchmarks: modified OSY SCVR HCR
Metrics: pareto hypervolume (PHV) maximum discovered feasible preferred objective value
Baselines: PAC-MOO-0 USEMOC evolutionary baselines
Builds on: MESMOC USEMOC
Closest papers by meaning
- Multi-Objective Bayesian Optimization with Active Preference Learning (AAAI 2024 · similarity 0.92)
- User Preference Meets Pareto-Optimality in Multi-Objective Bayesian Optimization (AAAI 2025 · similarity 0.91)
- FERERO: A Flexible Framework for Preference-Guided Multi-Objective Learning (NeurIPS 2024 · similarity 0.89)
- MOBO-OSD: Batch Multi-Objective Bayesian Optimization via Orthogonal Search Directions (NeurIPS 2025 · similarity 0.89)
- Principled Preferential Bayesian Optimization (ICML 2024 · similarity 0.88)
- Pareto Front-Diverse Batch Multi-Objective Bayesian Optimization (AAAI 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.