Amortized Active Generation of Pareto Sets

We propose active generation of Pareto sets (A-GPS), a framework for online discrete black-box multi-objective optimization that learns a generative model of the Pareto set while supporting a-posteriori preference conditioning. A-GPS avoids costly hyper-volume computations and enables flexible sampling across the Pareto front without retraining. Experiments on synthetic functions and protein design tasks show strong sample efficiency and effective preference incorporation.