Permutation-equivariant deep reinforcement learning optimizes mobile edge computing

This research introduces a permutation-equivariant deep reinforcement learning approach to manage resources and service migration in mobile edge computing. The method treats resource allocation and service migration as a single constrained Markov decision process, solved using an enhanced deep deterministic policy gradient algorithm. The architecture efficiently scales with the number of users by sharing a per-user head that incorporates individual and system-level features.

Originally published by gnews