On the 24th, the reporter learned from the University of science and technology of China that researcher Li Zefeng of the University used machine learning methods to summarize the focal time function characteristics of more than 3000 earthquakes with m ≥ 5.5 in the world, panoramically display the similarity and diversity of global earthquake rupture processes, deepen the understanding of seismic energy release modes, and have enlightenment significance for early earthquake warning. The research results were recently published in the geophysical research express, an internationally renowned geoscience journal. Earthquakes are one of the important natural disasters facing human society. In the past 20 years, medium and large earthquakes around the world have caused nearly 1 million casualties and countless economic losses. There are many kinds of earthquake rupture processes. Objectively measuring their similarities and differences is helpful to understand the seismic physical process and the early prediction of earthquake magnitude. However, previous studies have either superimposed the average rupture process of multiple earthquakes, which can not measure the range of global earthquake differences, or based on the statistics of some rupture characteristics, it is impossible to make a systematic comparison of the whole rupture process. Researcher Li Zefeng used the variational self coder in depth learning to compress and reconstruct the focal time functions of more than 3000 medium and large earthquakes in two dimensions, and showed the global seismic moment release pattern and quantity distribution in a panoramic way. It is found that medium and large earthquakes are dominated by simple fractures and less complex fractures, and reveals the distribution law of two types of special earthquakes, that is, the escape mode in which the energy release is concentrated in the later stage of fracture and the complex earthquake with multiple energy release. It is found that the energy release mode of large earthquakes is weakly magnitude dependent, which provides a useful enlightenment for the predictability of the final magnitude in the early warning of earthquakes. This research achievement is the development of the source time function clustering method jointly studied by the team and Harvard University. It is also one of the series of research achievements that the team is committed to applying artificial intelligence to scientific discovery in recent years. (Xinhua News Agency)