Upcoming Oral Candidacy Exams
|Start:||8/12/2014 at 9:00AM|
|End:||8/12/2014 at 5:00PM|
|Location:||100 Stinson Remick and 117-I Cushing Hall|
August 12, 2014, 12:30 pm, 100 Stinson Remick
Advisor: Dr. Patrick Flynn
Committee Members: Dr. Kevin Bowyer, Dr. Laurel Riek, Dr. Aaron Striegel
“Detection of Critical Crowd Conditions in Pedestrian Flow”
Crowd behavior analysis, a subdomain of human activity recognition, is a broad topic in computer vision that includes crowd management, public space design, virtual environments, visual surveillance, and intelligent environments. Recent areas of high interest include interactions between a small group of people, such as gangs in a prison, and large crowds of people at gatherings such as a sporting venue, concert, or religious pilgrimage.
Tragic incidents, such as the stampeding that occurred at the 2010 Loveparade concert and the multiple incidents that have occurred at the Jamarat Bridge necessitate a need for crowd analysis. Since 1994, over 1000 people have been killed or injured in stampedes and related incidents during the Hajj at the Jamarat Bridge. In 2006, the bridge was redesigned to eliminate many of the bottlenecks and increase crowd flow. Since this time, there have been no major incidents.
Unfortunately for the victims of the Jamarat Bridge and many other stampeding incidents, crowd analysis was used to make improvements and insure the safety of individuals of future events. The proposed research of this dissertation project is designed for proactive measures. Empirical work has shown patterns in crowd flow, known as critical crowd conditions, which appear prior to stampeding. We propose a method to detect these critical conditions to provide proactive measures of assistance.
August 12, 2014, 3:00 pm, 117-I Cushing Hall
Advisor: Dr. Nitesh Chawal
Committee Members: Dr. Kevin Bowyer, Dr. Sidney D'Mello, Dr. Dong Wang
"Data Science for Imbalanced and Partially Labeled Datasets"
Data science is host to several challenging problems. Two of the most pervasive are the problems of class imbalance and class uncertainty, both of which are endemic to classification tasks. If we consider a class to be a categorization of an instance or observation, then the class imbalance problem is said to exist in a dataset if the occurrence rates for each class are not(roughly) equivalent, resulting in a difficult classification problem. This situation is often found in real-world data describing infrequent but important observations. A further complication frequently found in real-world data is that, despite their vital importance to the task of classification, class memberships or labels can be difficult to define and capture. Unlike the binary-class domain, where all instances are labeled, in the partially labeled domain some of the instance labelings are unknown. Leveraging both labeled and unlabeled instances is often desirable, as learning performance potentially improves with more instances. However, the addition of unlabeled data can make the underlying labeling function more difficult to discern. Accordingly, partially labeled environments further complicate imbalance learning, as one must determine a class boundary with instances of