In this talk, I will present novel algorithms addressing these challenges in Biology and Sustainability. First, I will describe Markov Chain Monte Carlo and variational inference techniques for integrating heterogeneous sources of information via multiple kernel learning. The motivation behind this work came from a challenging protein folding problem that is described by multiple information sources. Next, I will present sparse Bayesian models and Expectation-Maximization algorithms that scale up to continent-wide spatial inference with remote sensing data, and provide state-of-the-art estimates for future vegetation cover in the Arctic under climate change scenarios. Finally, I will discuss my most recent work on adaptive spatiotemporal exploratory inference for hemisphere-wide species distributions using remote sensing and massively crowdsourced (2.5M) eBird observations.
BIO: Theodoros (Theo) Damoulas is a Research Associate in the Department of Computer Science at Cornell University and a member of the Institute for Computational Sustainability. His main research interests are probabilistic machine learning, Bayesian statistics, spatiotemporal inference and kernel methods applied to biology and sustainability. He is also conducting research in the areas of bioacoustics, reinforcement learning, and human sensing.
Theo received his Ph.D. in computer science from the University of Glasgow in 2009, his M.Sc in computer science (Distinction) from the University of Edinburgh in 2005 and his M.Eng in mechanical engineering (1st Class) from the University of Manchester in 2003. He has received best paper awards at IEEE ICMLA and IAAI, he holds a US patent on automatic counterfeit banknote detection, and was awarded the NCR PhD fellowship (2006-2009) and subsequently the Classification Society Distinguished Dissertation award (2012) for his Ph.D thesis.
Refreshments will be provided at 10:30AM