arXiv:2510.21052v3 Announce Type: replace-cross 
Abstract: We introduce active generation of Pareto sets (A-GPS), a new framework for online discrete black-box multi-objective optimization (MOO). A-GPS learns a generative model of the Pareto set that supports a-posteriori conditioning on user preferences. The method employs a class probability estimator (CPE) to predict non-dominance relations and to condition the generative model toward high-performing regions of the search space. We also show that this non-dominance CPE implicitly estimates the probability of hypervolume improvement (PHVI). To incorporate subjective trade-offs, A-GPS introduces preference direction vectors that encode user-specified preferences in objective space. At each iteration, the model is updated using both Pareto membership and alignment with these preference directions, producing an amortized generative model capable of sampling across the Pareto front without retraining. The result is a simple yet powerful approach that achieves high-quality Pareto set approximations, avoids explicit hypervolume computation, and flexibly captures user preferences. Empirical results on synthetic benchmarks and protein design tasks demonstrate strong sample efficiency and effective preference incorporation.

تم تقديم إطار جديد لتحسين الأهداف المتعددة في الوقت الفعلي، يسمى التوليد النشط للأطقم الباريتو (A-GPS). تستخدم هذه الطريقة نموذجًا توليديًا لتعلم وتوقع علاقات عدم الهيمنة بناءً على تفضيلات المستخدم، مما يسمح بأخذ عينات فعالة عبر جبهة باريتو دون الحاجة إلى إعادة التدريب.

Se ha introducido un nuevo marco para la optimización multiobjetivo discreta en línea, denominado Generación Activa Amortizada de Conjuntos de Pareto (A-GPS). Este método utiliza un modelo generativo para aprender y predecir relaciones de no dominancia basadas en las preferencias del usuario, permitiendo un muestreo eficiente a través del frente de Pareto sin necesidad de reentrenamiento.

Un nouveau cadre pour l'optimisation multi-objectifs discrète en ligne, appelé Génération Active Amortie des Ensembles de Pareto (A-GPS), a été introduit. Cette méthode utilise un modèle génératif pour apprendre et prédire les relations de non-domination en fonction des préférences des utilisateurs, permettant un échantillonnage efficace à travers le front de Pareto sans nécessiter de réentraînement.

A new framework for online discrete black-box multi-objective optimization, called Amortized Active Generation of Pareto Sets (A-GPS), has been introduced. This method utilizes a generative model to learn and predict non-dominance relations based on user preferences, allowing for efficient sampling across the Pareto front without the need for retraining.

Amortized Active Generation of Pareto Sets

One More Thing in AI – Your Shortcut to AI Mastery

Amortized Active Generation of Pareto Sets

Was this article worth reading? Share it

One More Thing in AI

Gonogo by Trampoline

Adgrow

GoAgentic

Octofy

Guapa AI

Ready to build your own newsroom?