This month we have William M. Pottenger presenting "To Be or Not To Be IID: That is the question." William's abstract and bio are below.
Abstract: Much prior work has shown the practical value of modeling random variables as IID in order to simplify statistical inference, yet prior work has also shown this assumption to be suboptimal in terms of model performance. Occam’s razor prompts us to simplify explanations, and this talk will present how a very simple transform has been leveraged to improve performance of both generative and discriminative learners, as well as unsupervised learning, in a number of application domains including differentially private community discovery.
Bio: William M. Pottenger is an Associate Research Professor at Rutgers University at DIMACS and RUTCOR in the field of Computer Science. Bill is also Director of Transition for CCICADA, the DHS Command, Control and Interoperability Center for Advanced Data Analysis. He is also founder of Intuidex, a manufacturer of solutions in the visual and data analytics space. Bill is active in research and development of technology, and as principle investigator has received over $7M in competitive research funding from the NSF, DHS, NIJ, DOD, industry, etc. and over $30M as co-investigator. He has over 40 peer-reviewed publications, has served as editor and chair of several proceedings/symposia and made over 50 professional presentations/seminars. Bill is a member of ACM, IEEE, SIAM and has served as a program committee member/referee for numerous professional venues, journals, etc. Among other awards he is the recipient of the PC. Rossin Endowed Assistant Professorship and a United States Air Force Certificate of Appreciation. Prior to coming to Rutgers, Bill completed his Ph.D. in Computer Science at the University of Illinois at Urbana-Champaign and worked as a Research Scientist at the National Center for Supercomputing Applications and at Lehigh University. Bill’s research interests include the fields of statistical relational learning and information extraction as applied in Higher Order Learning™, a framework he developed for both supervised and unsupervised learning based on higher-order relations for which four patents have been allowed to date. He is active in research in visual and data analytics and parallel and distributed computing as well. His company, Intuidex, Inc., creates leading-edge visual and data analytics technology for use in business and government. Application domains of Intuidex technology include business, public safety, intelligence and defense. See www.intuidex.com (http://www.intuidex.com/) for more info on Intuidex’s Watchman Analytics™ , or email [masked] .