Earthquake declustering is a procedure used for a wide range of problems in seismology, and there are many different methods for it. As Deep Learning has been very successful at replicating or even exceeding many tasks in seismology, we wanted to try to use it for declustering as well. In order to achieve this, we created a large suite of synthetic catalogs using the Epidemic Type Aftershock Sequence model (ETAS) in which an event is labeled as a mainshock or an aftershock. We then trained a LSTM models for mainshock and afterschock classification,  where the best performing model achieved an accuracy of  85%.  

Atomistic computer simulations have an increasing role in biophysics as they provide spatial and temporal resolution. However, most biophysical events are rare and would take too long to simulate with classical molecular dynamics simulations. To overcome these limitations, we developed a strategy that joins the forces of artificial intelligence and parallel computing: AI for molecular mechanism discovery, or AIMMD. We benchmarked our method on the folding/unfolding of the mini-protein chignolin. To get the most out of limited resources, we split the computational burden into many parallel short simulations orchestrated by AI. Then, we combine them after inferring their statistical properties and approximate the result of long, unbiased simulations. The final result gives us accurate thermodynamic and kinetic information. AIMMD bears promising application in the investigation of metabolic pathways involved in the prevention of human diseases.