Brian D. Ziebart
Machine Learning PhD Candidate
Machine Learning Department
School of Computer Science
Carnegie Mellon University

I am a PhD student in the Machine Learning Department (part of the School of Computer Science) at Carnegie Mellon University. I work with Anind Dey and Drew Bagnell.

Interests

I am interested in building better models of structured human behavior. Two examples are the selection of driving routes and the scheduling and execution of activities in the home. With better models of these behaviors, computer systems can detect abnormal behavior, offer suggestions for improvement, adapt to anticipated needs, and generally make our lives a little better.

Manually specifying rules for modeling these behaviors is difficult, so we rely on machine learning techniques to construct models. We use examples of "good" behavior to build models that we then use for recommendation and anomaly detection, such as providing the best context-sensitive driving route or detecting when an elderly relative is not functioning normally.

Projects

Driver Modeling
Creating a better driving route recommendation system...
Part of the Quality of Life Technologies Center

Bayes Net Structure Learning
Expanding the class of structures where MAP estimation and Bayesian Model Averaging can be performed efficiently for Bayesian Network structure learning. For temporal models (DBNs), this allows inter-temporal and intra-temporal dependency modeling. Applied to Bayes Net classification (Augmented Naive Bayes classification), we can learn the Selective Forest-Augmented Naive Bayes (SFAN) classifier, which provides built-in feature selection for Tree-Augmented Naive Bayes (TAN) classifier with the same asymptotic time complexity.

Activity Recognition
I spent a summer at Intel Research Seattle with Matthai Philipose working on recognizing the activities of people in a sensored environment. This is very challenging because activities are governed by very complex rules (physical, common sense, etc) and can be interrupted, interleaved, or abandoned, making learning (and inference) hard.

Automating Ubiquitous Computing Environments
In my past life at the University of Illinois, I worked with Dan Roth and Roy Campbell on using machine learning algorithms to automate a computer-controlled environment based on user-demonstration. I started my research life in systems research.

Publications

B. D. Ziebart, A. K. Dey, and J. A. Bagnell, "Learning Selectively Conditioned Forest Structures with Applications to DBNs and Classification," in Uncertainty in Artificial Intelligence (UAI 2007), Vancouver, BC, July 2007. (Master's degree paper) [pdf]

B. D. Ziebart, D. Roth, R. H. Campbell, and A. K. Dey, "Learning Automation Policies for Pervasive Computing Environments," in IEEE International Conference on Autonomic Computing (ICAC 2005), Seattle, WA, June 2005.

A. Ranganathan, J. Al-Muhtadi, J. Biehl, B. Ziebart, R. H. Campbell, and B. Bailey, "Towards a Pervasive Computing Benchmark," in PerWare '05 Workshop on Support for Pervasive Computing at the Third IEEE International Conference on Pervasive Computing and Communications (PerCom 2005), Kauai Island, HI, Mar. 2005.

B. Ziebart and D. Roth, "Learning Context-Dependent User Preferences in a Ubiquitous Computing Environment," Undergraduate Thesis. Department of Electrical and Computer Engineering. University of Illinois at Urbana-Champaign. May 2004.

M. Roman, J. Al-Muhtadi, B. Ziebart, and R. H. Campbell, "System Support for Rapid Ubiquitous Computing Application Development and Evaluation," in Systems Support for Ubiquitous Computing Workshop, at the Fifth Annual Conference on Ubiquitous Computing (UbiComp 2003) , Seattle, Washington, Oct. 2003. [pdf]

M. Roman, B. Ziebart, and R. H. Campbell, "Dynamic Application Composition: Customizing the Behavior of an Active Space," in IEEE International Conference on Pervasive Computing and Communications (PerCom 2003), Dallas-Fort Worth, Texas, Mar. 2003. [pdf]

(Note: Some of this material is Copyright IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE)