Thursday 16 Feb 2017: Learning in Non-Geometric Spaces, Analysis of Protein Molecules, and Beyond
Dr. Lorenzo Livi - University of Exeter
Harrison 170 14:30-15:30
In this presentation, I will discuss about past, current, and future research topics that characterize my research activity. First, I will talk about learning in “non-geometric spaces”. Typical learning problems are conceived on geometric domains, such as Euclidean spaces. However, many real-world applications deal with data that are better characterized by more complex representations, such as labeled graphs and sequences. Here, I will draw a connection also with information-theoretic learning. Research on protein molecules provides a very important application domain where such methods play an important role. In fact, folded proteins can be suitably represented as graphs, which are typically called protein contact networks. I will elaborate over recent research results in the context of protein contact networks that offer a link with the field of complex systems. Notably, I will discuss about anomalous diffusion and generative models of protein contact networks. Finally, I will present current and future research directions pointing to recurrent neural networks, time-varying graphs, unsupervised learning, protein contact networks, and several aspects in computational neuroscience.