Multi‑objective optimization must accommodate diverse preferences over conflicting goals. Conventional approaches return a limited set of trade‑off solutions, whereas preference‑conditioned Pareto set learning maps a given preference directly to its corresponding Pareto‑optimal solution. In many real‑world settings only a small offline dataset is available, leaving the decision and objective spaces sparsely covered. Existing methods either optimize each preference separately or train a problem‑specific Pareto solution model, requiring repeated search across preferences or repeated training across problems.
We introduce In‑Context Preference‑Conditioned Pareto Set Learning (ICPSL), a framework that amortizes both preference‑wise solution search and problem‑specific model construction. ICPSL first pretrains a foundation‑style hypernetwork on a large, diverse collection of synthetic prior data. Given offline observations of an unseen problem, the pretrained hypernetwork generates a problem‑specific, preference‑conditioned Pareto solution model in a single forward pass. The resulting model can then predict solutions for arbitrary preferences without further training or search.
Experiments on synthetic and real‑world tasks show that ICPSL produces high‑quality Pareto solutions while delivering substantial speedups in constructing problem‑specific solution models.
Review