Project Description: From violent and explosive volcanic eruptions, to quiet lava flow effusions, volcanic activity covers a wide spectrum. Generally, volcanoes can be categorized as either: active, dormant, or extinct. However, the classification of “active” is a bit of a misnomer as volcanoes often go through intermittent periods of variable thermal activity. There exists a general consensus regarding the underlying mechanisms involved in volcanic eruptions, however, a problem which has plagued geophysicists is: how do we predict the waxing and weaning of volcanic activity? This is an essential step towards combating volcanic hazards. This project will use existing data to forecast volcanic behavior with machine learning methods in statistical learning theory.
Deep learning has had significant recent success, and is outperforming other standard machine learning methods (such as support vector and other kernel machines) on a variety of tasks, including time series prediction. We wish to compare the results produced by the deep convolutional neural network approach, with more principle techniques that are well grounded in statistics and information theory. Techniques include, Bayesian inference, distributional clustering, and Gaussian Processes.
Date: January 2019
Mentor: Dr. Susanne Still
Project Funding:
- Undergraduate Research Opportunities Program (UROP) Grant