The RF machine learning research is mainly aimed at complex electromagnetic environments, studying and exploring the paradigm of building large data sets and designing large models suitable for electromagnetic signal processing. At the same time, we try to integrate expert knowledge such as system theoretical frameworks, prior statistical information and classical signal processing methods with AI algorithms, and study AI algorithms jointly driven by "model + data" to achieve deep cognition and optimal decision-making of the electromagnetic environment, and serve the next generation of wireless systems.