1:35 PM - 1:55 PM
[S21-10] [Invited]Potential pathway to understanding earthquakes using a data-driven approach: learning from other informatics applications
The quest to comprehend the complex and intricate mechanisms underlying large magnitude earthquakes has been a central pursuit in the field of seismology. In recent years, the advent of data science, coupled with the emergence of the fourth paradigm of science, has provided a novel avenue to revolutionize our understanding of seismic events. This talk explores the potential of harnessing data-driven methodologies, inspired by informatics applications from diverse scientific domains, to unravel the enigmatic nature of earthquakes.
The fourth paradigm of science emphasizes the significance of data-intensive research in shaping the trajectory of scientific discovery. Historically, seismology has been constrained by the limitations of traditional observational techniques, which often fail to capture the multifaceted dynamics of earthquake generation. However, the integration of modern data science techniques offers an unprecedented opportunity to gather, process, and analyze vast datasets to extract meaningful insights. This paradigm shift enables researchers to move beyond empirical observations and delve into the realm of data-driven hypothesis generation, refining our comprehension of the underlying processes that lead to seismic activity.
The potential of the modern data-driven approach is apparent, as the first two paragraphs of this abstract were actually written by ChatGPT, the famous chatbot service based on a large language model trained by deep learning technology. Recent research has demonstrated that deep learning can significantly improve the efficiency and accuracy of processing seismic data into information useful for further analyses, or can assist different tasks in seismic monitoring. However, cautions shall be made when using such convenient tools as the “no free lunch theorem” holds here as well. Users are required to correctly interpret the prediction results from a machine learning model under the well-known limitation of such data-driven models being effective only within the domain of their training data.
The inherent complexities of earthquake systems demand careful consideration of data quality, feature selection, and model interpretability. In this talk, I will draw inspirations from informatics applications in other scientific disciplines, such as materials science, and discuss potential strategies that seismologists can further capitalize on the wealth of seismic data generated from ground-based sensors, satellite observations, and historical records. These strategies are founded on some recent understandings of the learning mechanisms of deep learning, and the successful applications of transfer learning technologies in various science and engineering problems with limited data.
The goal of this talk is to explore the possible roles of data science in making a bridge between theoretical seismology and data-driven insights. As we stand on the cusp of this interdisciplinary revolution, collaborative efforts between seismologists and data scientists are poised to reshape the landscape of seismic research and disaster preparedness.
The fourth paradigm of science emphasizes the significance of data-intensive research in shaping the trajectory of scientific discovery. Historically, seismology has been constrained by the limitations of traditional observational techniques, which often fail to capture the multifaceted dynamics of earthquake generation. However, the integration of modern data science techniques offers an unprecedented opportunity to gather, process, and analyze vast datasets to extract meaningful insights. This paradigm shift enables researchers to move beyond empirical observations and delve into the realm of data-driven hypothesis generation, refining our comprehension of the underlying processes that lead to seismic activity.
The potential of the modern data-driven approach is apparent, as the first two paragraphs of this abstract were actually written by ChatGPT, the famous chatbot service based on a large language model trained by deep learning technology. Recent research has demonstrated that deep learning can significantly improve the efficiency and accuracy of processing seismic data into information useful for further analyses, or can assist different tasks in seismic monitoring. However, cautions shall be made when using such convenient tools as the “no free lunch theorem” holds here as well. Users are required to correctly interpret the prediction results from a machine learning model under the well-known limitation of such data-driven models being effective only within the domain of their training data.
The inherent complexities of earthquake systems demand careful consideration of data quality, feature selection, and model interpretability. In this talk, I will draw inspirations from informatics applications in other scientific disciplines, such as materials science, and discuss potential strategies that seismologists can further capitalize on the wealth of seismic data generated from ground-based sensors, satellite observations, and historical records. These strategies are founded on some recent understandings of the learning mechanisms of deep learning, and the successful applications of transfer learning technologies in various science and engineering problems with limited data.
The goal of this talk is to explore the possible roles of data science in making a bridge between theoretical seismology and data-driven insights. As we stand on the cusp of this interdisciplinary revolution, collaborative efforts between seismologists and data scientists are poised to reshape the landscape of seismic research and disaster preparedness.