Abstract

This talk presents our recent research on large language models for telecommunications, with the goal of extending foundation models into the telecom domain. Our motivation stems from the observation that general-purpose LLMs lack layer information such as radio-frequency (RF) signals. The first part introduces TelecomGPT, the first large language model tailored for the telecom domain. It covers the full pipeline from data processing and model training to evaluation and structured understanding and reasoning over technical standards and domain-specific data. The second part presents RF-GPT, the first large language model designed for RF language alignment mechanisms, enabling the model to understand and reason over essential priors for RF signal understanding, and this work provides an initial step toward addressing this gap. Overall, this line of research points toward a new paradigm of AI-native networks, where foundation models enable unified representations, cross-task reasoning, and deep integration of AI into wireless systems.

Biography

Dr. Hang Zou is currently a Postdoctoral Fellow at the Digital Future Institute, Khalifa University, working with Prof. Merouane Debbah. He pioneered the concept of the radio-frequency language model (RFLM) and developed the first such model, RF-GPT. He received his Ph.D. in Wireless Communications from Paris-Saclay University in 2022, under the supervision of Prof. Samson Lasaulce. Prior to joining Khalifa University, he was a Researcher at the Technology Innovation Institute, where he worked across the full LLM stack, from data pipelines and training to inference and evaluation, and developed TelecomGPT, the first telecom-specific LLMs, as well as the Falcon-Edge family of efficient language models. His research interests include 6G, semantic and goal-oriented communications, telecom-specific LLMs, edge AI, model compression, and AI-driven optimization. He has authored and co-authored publications in leading venues such as Nature Reviews Electrical Engineering and IEEE JSAC, and actively serves as a reviewer for major IEEE journals and conferences.