With a new generation of artificial intelligence around the corner, artificial intelligence (AI) has played an increasingly important role at many levels and fields. In recent years, AI technologies have developed rapidly, and related algorithms have successfully solved problems in many fields, such as unmanned driving, smart home, medical and healthcare technology, and intelligent security. At the same time, AI algorithms have also opened a door for geophysicists to find more efficient solutions to complex problems and to obtain new geophysical insights from large amounts of data. For example, geophysicists no longer need to rely exclusively on numerical solutions of the partial differential equations that describe geophysical systems. Instead, machine learning techniques often allow to learn general solutions from training data, which in turn may allow orders-of-magnitude more efficient computations, such as for inverse problems often encountered in geophysical exploration. Moreover, AI methods like unsupervised learning, data mining, and deep learning such as artificial neural networks, enable geophysicists to analyze vast amounts of data in an automated fashion and to obtain new insights into the behavior of complex geophysical systems that are not intuitively obvious. The direction of "Artificial Intelligence and Interdisciplinary Research" will be based on the interdisciplinary innovations of geophysics, applied mathematics, and computer science. Research will be conducted on AI as applied to geoscience-related problems, their generalization and interpretability, and the dynamic coupling of multi-source data. Particular emphasis will be on solving problems commonly encountered in geophysics, such as insufficient sampling, diverse data formats, data heterogeneity and imbalance, and to use AI approaches to obtain a better understanding of the Earth. The specific contents involve deep learning, transfer learning, knowledge graphs, federated learning, and their applications on oil and gas exploration, digital signal processing, deep earth imaging, remote sensing, microseism identification, deep time of the Earth, data assimilation. Interested faculty includes the following.

Ma, Jianwei

　　Main interest includes dictionary learning, deep learning, and applications of artificial intelligence on seismic signal processing, inversion, image processing, and data assimilation.

Hu, Tianyue

　　Main interest includes artificial-intelligence machine learning based explorational seismic data acquisition, processing technology, oil and gas reservoir identification, and so on.

Yue, Han

　　One of his research interest is to realize automatic seismic event detection using deep learning algorithms. In comparison with tradition model driven detection methods, deep learning algorithms are more applicable to problems that are difficult to describe mathematically and easy to master in experience. Thus the deep learning detection algorithms generally have better performance than traditional algorithms in image and voice detection problems. The seismic event detection problem is one of such a problem. By exploiting the parallel information of three-component seismograms and the causal relationship of seismogram timeseries, his group utilizes convolutional and recurrent neural networks to realize the automatic labeling of seismic/noise signals, which improves the performance of automatic seismic event detection.

Wang, Teng

　　Radar remote-sensing images contain rich information about the geometrical and physical properties of the earth surface. Using AI is an efficient way to reveal this information from large amount of radar images. Based on the imaging mechanism and application scenarios, we can simulate radar images and interferograms, superimposed with multi-dimensional atmospheric delays and random noises. In this way, we can form sufficient, diverse, high-quality, labeled samples to train the deep learning network, significantly improving the accuracy and efficiency of digging information from remote-sensing images. Application fields include phase unwrapping, mask generation, atmospheric estimation, local deformation detection, post-earthquake damage evaluation, change detection and many more.

Song, Xiaodong

　　Main interest is on the applications of AI and machine learning to seismology and the physics of the Earth's interior, including seismic signal processing, geophysical inversion of the internal structure of the earth, processing of large volumes of seismic data, and detection and extraction of seismic temporal change signals.

Berndt, Thomas A.

　　Paleomagnetism provides valuable insight into the past of the Earth through magnetic records in rocks, but it encounters heterogeneities at various different scales: From the differences of one magnetic particle to the other at nanometer scale, over the differences in lithologies of one rock sample to the next, to the changes in tectonic, environmental and geomagnetic factors that impact magnetic signals over thousands of kilometers and millions of years. Currently, the field largely relies on the experience of paleomagnetists to interpret magnetic data preserved in rocks, each in their unique setting. Machine Learning techniques are a promising tool to integrate data on all these scales, learn and interpret underlying patterns from data rather than from experience.