Next generation of machine learning and meta-knowledge creation from data
Presented by Dr. Dimitris MetaxasRutgers University
June 15, 2017875 N. Randolph Street, Arlington, VA Room 40031:30 pm to 2:30 pmAbstract: A central goal of artificial intelligence is the production of autonomous agents that are able to interact with, learn from, and reason about complex, real-world environments and humans in the absence of predefined knowledge. Existing autonomous systems currently depend largely on predefined, human-supplied meta-knowledge about the world. We propose a novel dynamic data-driven approach to meta-knowledge creation for complex learning and meta-reasoning. Our approach is centered on the following problems: (1) how to autonomously construct and update rich knowledge bases from sensory data in real-world scenarios, (2) how to use this knowledge to reason about and create agents that successfully interact with the environment, and (3) how to identify when data-driven knowledge and reasoning capabilities are insufficient and, in response, learn how to augment and fix the system’s cognitive facilities. We first introduce a new algorithm for deriving higher-order semantic information from a set of primitive semantic concepts. Our algorithm is based on non-negative matrix factorization, and we call the resulting derived semantic representation the "scenario". We extend the proposed algorithm to construct hierarchies of scenarios. These hierarchies enable us to perform complex reasoning tasks in an efficient manner. Such tasks include efficient object search, improved object detection, and complex visual querying. Finally, we discuss future work on how to actively learn and utilize scenarios in dynamic environments.
Bio: Dr. Dimitris Metaxas is a Distinguished Professor and Chair of the Computer Science Department at Rutgers University. He is director of the Center for Computational Biomedicine, Imaging and Modeling (CBIM). From September 1992 to September 2001 he was a tenured faculty member in the Computer and Information Science Department of the University of Pennsylvania and Director of the VAST Lab. Dr. Metaxas received a Diploma in Electrical Engineering from the National Technical University of Athens, Greece, in 1986, an M.Sc. in Computer Science from the University of Maryland, College Park, in 1988, and a Ph.D. in Computer Science from the University of Toronto in 1992. Dr. Metaxas has been conducting research towards the development of formal methods to advance medical imaging, computer vision, computer graphics, and understanding of multimodal aspects of human language. His research emphasizes the development of formal models in shape representation, deterministic and statistical object modeling and tracking, sparse learning methods for segmentation and restoration, and organ motion analysis. Dr. Metaxas has published over 500 research articles in these areas and has graduated 45 Ph.D. students. The above research has been funded by NSF, NIH, ONR, AFOSR, DARPA, HSARPA, and ARO. Dr. Metaxas has received several best paper awards, and he has 8 patents. He was awarded a Fulbright Fellowship in 1986, is a recipient of an NSF Research Initiation and Career awards, an ONR YIP, and is a Fellow of the MICCAI Society, the American Institute of Medical and Biological Engineers, the IEEE. He has been involved with the organization of several major conferences in vision and medical image analysis, including ICCV 2007, ICCV 2011, MICCAI 2008, and CVPR 2014.
This lecture is presented by invitation only. Please contact Dr. Tristan Nguyen for details. firstname.lastname@example.org; tel 703 696 7796