Monday April 6 4:00 - 4:50 PM Kelley 1001
Tomas Singliar Boeing Research and Technology Seattle, WA Learning to detect vehicular incidents from noisy data Many deployed traffic incident detection systems use algorithms that require significant manual tuning. We want to develop a learning detector to reduce the need for manual adjustments by taking advantage of a massive database of traffic sensor network measurements. First, we show that a rather straightforward supervised learner based on the SVM model outperforms a fixed detection model used by state-of-the-art traffic incident detectors. Second, we seek further improvements of learning performance by correcting misaligned incident times in the training data. The misalignment is due to an imperfect incident logging procedure. We propose a label realignment model based on a dynamic Bayesian network to re-estimate the correct position (time) of the incident in the data. Training on the automatically realigned data consistently leads to improved detection performance in the low false positive region. The data realignment framework requires inference in dynamic Bayesian networks. This is a computationally intensive process. However, for a class of networks naturally arising in detection and diagnostic problems - networks with a persistence property - a very efficient general inference algorithm exists. In the second part of the talk, I will describe this algorithm inspired by incident detection and other applications. Biography: Tomas has recently graduated from University of Pittsburgh and joined Boeing Research and Technology. His research interests in machine learning are inference and learning in Bayesian networks and scaling up learning technology to large real-world systems. He is also interested in anomaly detection and planning. He published his work in JMLR, UAI, AAAI and several other venues.
_______________________________________________ Colloquium mailing list [email protected] https://secure.engr.oregonstate.edu/mailman/listinfo/colloquium
