Workshop on Foundation Models, Generative AI, and Agentic Intelligence for AI-Native Radio Access Networks (AI-RAN)

» Workshop website «

Radio Access Networks (RANs) are evolving toward disaggregation, virtualization, and open ecosystems, enabling flexibility while increasing operational complexity. Classical model-driven optimization and rule-based control struggle with non-stationary propagation, traffic variability, service heterogeneity, and tightly coupled cross-layer interactions. These challenges motivate a shift from “AI as an add-on optimizer” to “AI as a first-class architectural component,” where learning, inference, and decision-making are embedded into the RAN while meeting real-time, reliability, and interoperability requirements. This workshop focuses on the architectural, algorithmic, and systemlevel foundations of AI-Native RAN, emphasizing how modern AI-such as foundation models, GenAI, LLMs, and agentic AI-can enable intentdriven operation, policy generation, knowledge reuse, and distributed coordination. We further aim to connect AI-native design with closedloop learning and RAN Intelligent Controller–enabled intelligence, while promoting reproducible research through datasets and prototype validation in open RAN environments.

Topics of interest include:

• AI-native RAN architectures and system design for disaggregated/open and cloud/edge-native RAN
• RIC-enabled intelligence and control frameworks (near-real-time (RT)/non-RT)
• Closed-loop learning-and-control for RAN
• Learning-driven radio resource management
• Foundation models/GenAI/LLM-enabled RAN operation
• LLM/GenAI for telecom knowledge engineering (retrieval-augmented generation (RAG)/knowledge graphs)
• Agentic and multi-agent intelligence for distributed RAN coordination
• Data-centric AI-RAN pipelines (telemetry, labeling/weak supervision, privacy-preserving data sharing, and continual learning)
• Digital twins/simulators/world models for RAN (sim-to-real transfer, synthetic data generation)
• Efficient AI for RAN and AI on RAN (edge deployment, acceleration/compression, task scheduling)
• Trustworthy and secure AI-RAN (attack detection/mitigation, explainability and traceability)
• Testbeds, datasets, over the air validation, and reproducible benchmarking