Reproduce “Online Frequency Scheduling by Learning Parallel Actions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anastasios Giovanidis, Mathieu Leconte, Sabrine Aroua, Tor Kvernvik, David Sandberg
Ericsson Research introduced novel Deep Q-Network architectures designed for online frequency scheduling in multi-user MIMO systems, which facilitate parallel decision-making for ultra-low inference latency. These architectures consistently surpassed traditional heuristic schedulers, delivered performance comparable to AlphaZero-based reinforcement learning at over 16 times the speed, and exhibited robust online adaptation to evolving wireless conditions.
Key insights
- Parallel decision-making architectures significantly reduce inference latency, making AI-driven frequency scheduling feasible for real-time 5G/6G requirements.
- Parameter-efficient designs like Unibranch and GNNs enable practical deployment of complex RL models by scaling to many sub-bands with reduced memory and computational overhead.
- The DQN framework, coupled with prioritized experience replay, naturally supports online fine-tuning and adaptation, effectively addressing the sim-to-real gap in dynamic wireless environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.