FedCritic: Serverless Federated Learning for 6G Resource Allocation
A recent study introduces FedCritic, a serverless federated multi-agent actor-critic framework designed for managing downlink resources in 6G ultra-dense networks. This framework tackles the challenges of joint subcarrier scheduling and power distribution while adhering to long-term per-user quality-of-service (QoS) minimum-rate requirements and interference coupling. In contrast to centralized training with decentralized execution (CTDE) methods, which depend on centralized critic training and combined trajectory aggregation, FedCritic utilizes lightweight gossip-based parameter averaging across the interference graph, allowing for reliable value estimation without a central server. The research emphasizes multi-cell orthogonal frequency-division multiple access (OFDMA) systems, where aggressive frequency reuse intensifies inter-cell interference (ICI). Virtual-queue deficit weights help maintain long-term QoS. The paper can be found on arXiv with the reference 2605.21418.
Key facts
- FedCritic is a serverless federated multi-agent actor-critic framework for 6G resource allocation.
- It addresses joint subcarrier scheduling and power allocation in multi-cell OFDMA systems.
- The framework uses gossip-based parameter averaging over the interference graph.
- It enforces long-term per-user QoS minimum-rate constraints via virtual-queue deficit weights.
- The approach avoids centralized training with decentralized execution (CTDE).
- The paper is published on arXiv with reference 2605.21418.
- It targets 6G ultra-dense networks with strong inter-cell interference.
- The research focuses on distributed downlink resource management.
Entities
Institutions
- arXiv