Trustworthy, interpretable and sustainable decentralized intelligence for 6G networks

An arXiv paper proposes a unified perspective on 6G decentralized intelligence, arguing that decentralization, trustworthiness, interpretability and sustainability must be designed jointly. Centralized processing of massive heterogeneous data brings communication overhead, latency bottlenecks and single points of failure, so decentralization becomes a fundamental architectural requirement for network intelligence and zero-touch operations. Security is a key enabler of this paradigm, but pushing intelligence to the edge via federated learning (FL) and decentralized federated learning (DFL) introduces complex trade-offs.