Thermal conductivity is one of the fundamental properties of materials, which is closely related to a wide range of scientific and industrial applications, including thermoelectric energy conversion, thermal barrier coatings, chip and battery thermal management, etc. Finding material systems with extreme thermal conductivity (extremely high or low) is an important prerequisite for achieving the aforementioned applications, which poses the main challenge in this field. At present, a comprehensive theoretical calculation method for thermal conductivity has been established based on first-principles calculations, but the actual calculation cost is huge, making it difficult to carry out exploration based on high-throughput calculations. On the other hand, in recent years, artificial intelligence-based material performance prediction methods have been developing rapidly and have also made achievements in thermal conductivity prediction. However, the existing methods are mainly limited by the insufficient number of computational and experimental thermal conductivity datasets, and are also lack of interpretability. In response to the above issues, a research team from Germany has developed a machine learning framework for screening materials with extreme thermal conductivity based on a combination of symbolic regression and sensitivity analysis technique. Compared to other machine learning methods, symbolic regression can not only achieve black-box like performance prediction, but also yield the analytical expression of the target performance, making it interpretable. Specifically, researchers conducted symbolic regression studies on the experimental thermal conductivity of 75 materials at 300K, and then obtained the functional relationship between thermal conductivity and descriptors. Further sensitivity analysis was conducted to obtain several descriptors that are most relevant to thermal conductivity, such as average mass, Debye temperature, and anharmonic parameters. Finally, based on the difficulty of calculating the above descriptors, a hierarchical screening on thermal conductivity can be carried out based on the inorganic material database. Based on the above framework, the researchers screened 732 material systems and found 96 thermal insulation materials with thermal conductivity below 10 W/mK. The main framework proposed in this study can not only be used for predicting thermal conductivity, but also have the potential to be used for high-throughput prediction of material properties with features similar to thermal conductivity, i.e. with limited existing databases, difficult to measure and with complex mechanisms.