Abstract
This paper studies a deep learning approach for multi-objective network optimizations. Heterogeneous performance measures are maximized simultaneously to identify complete Pareto-optimal tradeoffs. To this end, a multi-objective optimization (MOO) problem is first reformulated as a collection of constrained single objective optimization (SOO) problems, each associated with a Pareto-optimal point. A novel MOO learning mechanism is developed to address multiple instances of such SOO problems concurrently. A constrained optimization technique is parameterized with neural networks to find an individual solution of the Pareto boundary points. The developed scheme proves efficient in characterizing the optimal tradeoffs of conflicting performance metrics in interfering networks.