Megha Chakraborty gained her doctorate at Goethe University on 26 November. She uses deep learning as a tool to decode seismological data. Impressive publications and the award of the title „summa cum laude“ reflect her successful research.

Megha Chakraborty's cumulative dissertation consists of five research papers on the two major areas of seismology, earthquake monitoring and characterisation, and shear wave splitting analysis. Rapid and reliable characterisation of parameters such as time of occurrence, magnitude, and source properties from continuous recordings is used for early earthquake warning. Shear wave splitting analyses the underlying seismic structures in a region and provides valuable insights into the dynamic processes in the Earth's mantle; this area is still relatively unexplored by deep learning.

A significant outcome of her research is SAIPy, an open-source Python package that she created along with her colleague Wei Li. It provides a simple interface for the application of deep learning models. They were also able to identify
and estimate their magnitude,which is very important for early warning. The group demonstrated the applicability to the aftershocks of the 2023 Turkey earthquake.

The advanced system allows seismologists to use pre-trained models, retrain them on new data and monitor large amounts of continuous data. This system is already in use. An extension makes it possible to identify developing earthquakes before the seismic background noise and estimate their strength - which is very important for early warning. 

Another model uses a convolutional encoder architecture to detect the vertical motion of very early earthquake waves, which play a particularly important role in understanding the faulting mechanism behind smaller earthquakes. This deep learning model classifies the first waves automatically and efficiently, which significantly speeds up predictions. 

Chakraborty's latest publication describes a complex deep learning model that calculates certain parameters from waveforms that are crucial to the study of the internal structure and dynamic of the Earth’s interior. The comparison of this method with previously published results is very promising. It is the first attempt to use deep learning to decipher splitting parameters from waveforms and serves as an important basis for the applicability of such methods.

Chakraborty completed her doctorate in the Seismology and Artificial Intelligence research group of FIAS Fellow Nishtha Srivastava, who supervised the dissertation together with FIAS Fellow Georg Rümpker. Some of her work was presented to a wider public by FAZ and deutschland.de.

‘‘I am very grateful to my mentors, friends and colleagues at FIAS for their never-ending support in my phD journey," says Chakraborty. It has been a year since she started working at a Berlin-based company developing chatbots for insurance companies using large language models. "This has been a great learning experience for me and I am excited to witness what the future of AI beholds."

Publications:

• M. Chakraborty, W. Li, J. Faber, G. Rümpker, H. Stoecker and N. Srivastava, 2022. A study on the effect of input data length on a deep-learning-based magnitude classifier. Solid Earth, 13(11), pp.1721-1729.
• M. Chakraborty, C. Q. Cartaya, W. Li, J. Faber, G. Rümpker, H. Stoecker, N. Srivastava, PolarCAP – A deep learning approach for first motion polarity classification of earthquake waveforms, Artificial Intelligence in Geosciences, Volume 3, 2022a, Pages 46-52, ISSN 2666-5441, https://doi.org/10.1016/j.aiig.2022.08.001.
• M. Chakraborty, D. Fenner, W. Li, J. Faber, K. Zhou, G. Rümpker, et al. (2022b). CREIME—A Convolutional Recurrent model for Earthquake Identification and Magnitude Estimation. Journal of Geophysical Research: Solid Earth, 127, e2022JB024595. https://doi.org/10.1029/2022JB024595
• W. Li*, M. Chakraborty*, C. Q. Cartaya, J. Köhler, J. Faber, M. A. Meier, G. Rümpker, N. Srivastava, SAIPy: A Python package for single-station earthquake monitoring using deep learning, Computers & Geosciences, Volume 192, 2024, 105686, https://doi.org/10.1016/j.cageo.2024.105686.
• M. Chakraborty, G. Rümpker, W. Li, J. Faber, N. Srivastava, and F. Link, “Feasibility of Deep Learning in Shear Wave Splitting analysis using Synthetic-Data Training and Waveform Deconvolution”, Seismica, vol. 3, no. 1, Mar. 2024.