EdgeFed H-MAAC: Edge Federated Heterogeneous Multi-agent Actor-Critic

This repository contains a gym module for UAV-assisted MEC environment simulation and a TensorFlow implementation of EdgeFed H-MAAC framework.

Run

• To simulate the MEC systems in the paper, standard gym modules are implemented by MEC_env/mec_def.py and MEC_env/mec_env.py.
• An edge-federated actor-critic RL framework with mixed policies, abbreviated as EdgeFed H-MAAC, is developed in MAAC_agent.py.
• A mixed DDPG based algorithm AC_agent.py is also implemented as a baseline.
• Run *_run.py to test the algorithms in the simulated MEC system.

References

If you find the codes useful, please cite the following papers:

• Federated Multi-Agent Actor-Critic Learning for Age Sensitive Mobile Edge Computing [J]. IEEE Internet of Things Journal, 2021.

• An Edge Federated MARL Approach for Timeliness Maintenance in MEC Collaboration [C]//2021 IEEE International Conference on Communications Workshops (ICC Workshops). IEEE, 2021: 1-6.

---