A new machine learning model that generates realistic seismic waveforms will reduce manual labor and improve earthquake detection, according to a study recently published in JGR solid earth.
“To verify the effectiveness of our generative model, we applied it to seismic data collected in Oklahoma,” said Youzuo Lin, a computer scientist in the geophysics group at Los Alamos National Laboratory and principal investigator on the project. “Through a sequence of qualitative and quantitative tests and benchmarks, we have seen that our model can generate high-quality synthetic waveforms and improve machine learning-based earthquake detection algorithms.
Detecting earthquakes quickly and accurately can be a difficult task. Visual detection by people has long been considered the gold standard, but requires intensive manual work that does not adapt well to large data sets. In recent years, automatic detection methods based on machine learning have improved the accuracy and efficiency of data collection; however, the accuracy of these methods relies on having access to a large amount of high quality labeled training data, often tens of thousands of records or more.
To solve this data dilemma, the research team developed SeismoGen based on a Generative Antagonist Network (GAN), which is a type of deep generative model that can generate high-quality synthetic samples in several domains. In other words, deep generative models train machines to do things and create new data that might pass for real.
Once trained, the SeismoGen model is capable of producing realistic seismic waveforms of multiple labels. When applied to real-world earthquake data sets in Oklahoma, the team found that the increase in synthetic waveform data generated by SeismoGen could be used to improve the earthquake detection algorithms of land in cases where only small amounts of labeled training data are available.
Source of the story:
Material provided by DOE / Los Alamos National Laboratory. Note: Content can be changed for style and length.