My PhD thesis defense on few-shot multitask optimization. Many real-world problems allow only a handful of evaluations, which makes finding high-quality solutions hard, especially when a whole Pareto set must be learned. This thesis tackles the challenge through multitask optimization, developing model-based guidance frameworks that transfer knowledge across related tasks, and generalizing them from finite task collections to continuously parameterized task families in both single-objective and multiobjective settings.

Sep 24, 2026

Families of expensive multi-objective problems under varying operational conditions span a continuous task-parameter space with infinitely many multi-objective instances. We amortize the multi-objective optimization cost across that space by learning an inverse model, alternating between generative solution sampling and acquisition-driven search, so that solutions for unseen parameterized tasks can be predicted directly without expensive multi-objective re-evaluation.

Aug 20, 2026

Multitask optimization and sequential transfer are two tributaries of transfer optimization. We position that sequential transfer conducted in an iterative way can be used to solve multitask optimization. And a curated task prioritization approach can be used to elevate both multitask optimizer and sequential optimizer in multitask optimization problems.

Jun 22, 2026

How we generalize multi-task optimization to infinitely many-task optimization by learning and utilizing evolving task models.

Jun 6, 2026