Roni Khardon

Department of Computer Science
Tufts University

Research Interests:
Machine Learning and Data Mining
Artificial Intelligence
Theoretical Computer Science and Efficient Algorithms

Recent Projects

For Research, Publications, Datasets, Software, CV ... please see the Machine Learning Group Web Site .
See also the AI pages (under construction)

Machine Learning: Research Highlight: Time Series for Astronomy Computational Learning Theory: Research Highlight: On Line Learning Learning Planning and Acting in Complex Domains: Research Highlight: FODDs
Various projects are exploring foundations and interdisciplinary applications of machine learning, for astronomy, remote sensing, environmental engineering, and more. For astronomy we have investigated several approaches to signal processing, analysis, and classification of time series capturing the of luminosity of stars, with the goal of automating the categorization of light sources in astronomy sky surveys. To date we have focused on identifying variable periodic stars and on identifying quasars. In on-line learning the learner sees one example at a time, makes a prediction and updates his/her hypothesis but cannot access the entire data together. Our theoretical work has investigated convergence properties of linear separators in the on-line model in the context of learning Boolean functions. Most recently we proved convergence for on-line algorithms that aim to optimize the ranking of examples (a case applicable to ranking the results of search). In this line of research our goal is to develop theories and systems that will enable building agents that can act intelligently in complex structured non-deterministic environments. Our main effort in recent work has been to develop a knowledge representation (FODD) and algorithms for it that serve as the engine to build such agents. The work combines logical analysis, probabilistic inference, and algorithmic ideas, and solves planning problems at an abstract level. Recent application shows that our approach can solve difficult open world robot planning and be deployed on physical robots.
Publications: ECML 2009 / IJCAI 2009 Workshop / ECML 2010 / ApJ 2011 / ApJ 2012 / ECML 2012 / ApJ 2012 Publications: NIPS 2001/ COLT 2003/ JAIR 2005/ JMLR 2005/ ICML 2007/ JMLR 2007/ COLT 2012 Publications: IJCAI07 / UAI 2007 / JAIR 2008 / ICAPS 2008 / IJCAI 2009 / ICAPS 2010 / AAAI 2010 / JAIR 2011 / AIJ 2011 / ICRA 2012 / AAAI 2012 / ECML 2013 / NIPS 2013 / AAAI 2014 / AAAI 2015
To start exploring see this article: Astronomy Time Series To start exploring see this article: On-Line AUC Maximization To start exploring see this article: FODDs for Relational MDPs
See also videos from ICRA paper: simulation showing the communication between abstract planner and robot architecture and demo on simple physical robot.
Partly supported by NSF grant IIS-0803409. Partly supported by NSF grants IIS-0099446 and IIS-0803409. Partly supported by NSF grants IIS-0936687 and IIS 0964457.

For More Information:

Recent Talks

Recent Conference and Editorial Activities

Contact Information:

Roni Khardon
Department of Computer Science
Tufts University
161 College Ave.
Medford, MA 02155

Office: Halligan 230
Phone: 1-617-627-5290
Fax:     1-617-627-2227
Dept.: 1-617-627-2225