The 5th Prediction Science Seminar was held on 7th January 2022 delivered by Prof. Rose Yu from UC San Diego. The first part of the talk was about the integration of deep learning with spatiotemporal point processes to learn continuous space-time dynamics as in cases of earthquakes and COVID-19 spreads. The computation of the proposed method reaches up to 100 times faster than the standard approach with comparable or even better accuracy. The second part of the talk was focused on interactive neural processes with a new acquisition function called the latent information gain to efficiently train the model using optimal datasets. This approach has been implemented to speed up stochastic, spatiotemporal epidemic models in a real-world application.
For more information about the seminar series, please see the following page: