Priority-Constrained Descent for Hierarchical Multi-Objective Optimization
September 20, 2026
PCD is a gradient-based framework that exploits objective hierarchies by preserving the direction of primary descent. It allows for minimal distortion to satisfy secondary objectives, controlled by a single hyperparameter $\tau_{in}$.
HOW THIS AFFECTS YOU
●
researcherYou can better manage trade-offs between primary goals and constraints like sparsity or robustness.