Jonathan Warrell will give the talk, "Model Semantics in Deep Learning: Genomics, the Brain and Music Evolution."
Abstract: A gap has emerged in many domains between the most predictive models and models whose parameters are readily interpretable. This poses the challenge of how (and whether) we can extract knowledge from such optimally predictive models. I will discuss a general conceptual framework for analyzing issues related to interpretability in deep learning based on formal model semantics, particularly issues of generalization, relation to prior knowledge, and compressibility. I will then discuss how such considerations have motivated specific model architectures and analytic methods I have developed in confronting problems in a range of domains. These include developing integrated models of genetic risk for psychiatric disorders and cognition (including genetic, epigenetic, cellular and brain imaging data), detecting positive and negative selection in cancer, and identifying latent evolutionary processes in music. I will also discuss how techniques from statistical learning theory (PAC-Bayes) and dependent type theory can be used to provide a theoretical basis for the models I introduce.
Jonathan Warrell is a postdoctoral associate research scientist in the Computational Biology and Bioinformatics program at Yale University, working with Mark Gerstein. He has published extensively in computational biology, evolution, machine learning, computer vision, and music analysis. He is currently a member of several large-scale genomics consortia, including ENCODE, PsychENCODE, and PCAWG (Pan-Cancer Analysis of Whole Genomes), and his work has been featured in journals such as Science and Cell, as well as conferences such as CVPR, ECCV and ISMB. Jonathan began his academic career in music theory, with a BA from Cambridge in Music, and an MA and PhD from King's College London in Music Theory and Analysis. He then focused on machine learning and computer vision, with an MSc in Computer Science from University College London, followed by postdoctoral positions in computer vision at University College London and Oxford / Oxford Brookes Universities, and postdoctoral positions in computational biology at University of Cape Town and Yale University. His research areas include psychiatric genomics and cancer, interpretable machine learning, theoretical biology and evolution, statistical learning theory, and digital humanities (music).
Meeting ID: 859 1951 7074
Questions? Contact Lori Kelso