DeepCoMP: Multi-agent Reinforcement Learning for Multi-cell Selection in 5G and Beyond
summary
DeepCoMP is a working multi-agent deep RL system (PPO, OpenAI Gym) that learns dynamic multi‑cell selection from partial observations and a log‑QoE reward, implemented with Ray RLlib. Three variants are evaluated: centralized (best final QoE) and two distributed forms (D3Comp separate policies, DDComp shared policy). Distributed agents converge ≈200k vs >1M steps for centralized, and the method beats baselines by up to 231% in simulation; action limits and one connect/disconnect per step are enforced.
AI-written summary · about
chapters (8)
- 0:00Introduction to DeepCoMP
- 0:42Scenario and motivation
- 3:32Problem formulation
- 6:40Centralized DeepCoMP approach
- 9:04Distributed multi-agent approaches
- 10:26Implementation with Ray RLlib
- 18:05Evaluation results
- 22:17Conclusion and future work
related talks
30mLearning Multi-Agent Collaborations With Decomposition
11mUsing Deep Reinforcement Learning to Coordinate Multi-Modal Journey Planning with Limited ...
13mOne Protocol to Rule Them All: Wireless Network-on-Chip using Deep Reinforcement Learning
37mMultiagent Reinforcement Learning: Rollout and Policy Iteration
15mA Comparison of Reinforcement Learning Agents Applied to Traffic Signal Optimization
01h 17mTraining intelligent game agents using deep reinforcement learning
35mCrowd dynamics, RL and Unity: A Journey
31mFrom Single AI to Multi AI Systems