Earthquake prediction using deep learning
Seismology • AI •
This is one of my career-defining projects and one of the most memorable works I have done. While I was working as a researcher in Japan, I experienced many earthquakes that sparked my curiosity in seismology. As a Physics graduate, I found the lack of reliable earthquake prediction unsatisfying and wanted to find a robust solution. I started learning more about AI and seismology. That was also when TensorFlow was released, and interest in AI started growing. After much research, I developed a novel approach that can be an excellent precursor to big earthquakes.
Identifying foreshocks as and when they occur provides valuable insight into predicting imminent earthquakes. However, It was only possible to identify foreshocks after the entire seismic event had occurred. I created an algorithm based on deep learning that can discern foreshock, mainshock or aftershock in real time with high accuracy (>99%). The project is one of the first applications of deep learning in the field of seismology, and I am thrilled and excited about it and has the potential to save millions of lives.
I initially published my work in aRxiv and later presented my work at the annual seismological conference organised by SSA (Seattle, WA, 2019). I was also invited to present my talk at Google's ML Summit (Cambridge, MA, 2019)