Beckman Graduate Student Seminar Continues February 21

The Beckman Institute Graduate Student Seminar Series presents the work of outstanding graduate students working in Beckman research groups. The seminars are open to the UIUC campus. The next seminar takes place on Wednesday, Feb. 21, at noon in Room 1005 Beckman Institute.

Speakers and abstracts are listed below:

Counting the Nouns: Use of Simple Structural Cues in Early Verb Learning
Sylvia Yuan (CS)

How do children learn the meaning of new verbs? One influential proposal is that the kinds of sentence structure that verbs appear in help children with verb learning (Gleitman, 1990). Specifically, children start by using simple cues about the sentence structure, such as the number of nouns in the sentence, to learn something about the verb's meaning (Fisher et al., 1994). In a set of experiments, I examined whether and how such a simple cue can guide young children's interpretation of new verbs. Using a preferential-looking technique, we found that children two years and younger interpreted novel verbs differently depending on the number of nouns the verb is used with. Additional experiments explored how children use this cue, and whether they can retain these syntactic facts about the verb and retrieve them later to guide interpretation of the same verb used in a referential context. Our findings suggest that early on, children are sensitive to simple syntactic cues in interpreting new verbs, and that verb learning takes place as children gather these simple syntactic facts about the verb from their listening experience.

Robust Tracking by Fusion of Multiple Cues
Tony Xu Han (IFP)

Object tracking is the first step of many vision applications such as video surveillance, perceptual user interfaces, automated video content retrieval, and audio-visual speech analysis. It has drawn attention from the researchers in the vision community for at least two decades. In order to achieve robust tracking, the algorithm should overcome several difficulties, such as, ambiguity caused by a cluttered background and distracting objects, appearance variation and occlusion. It seems that none of the existing features/cues alone can achieve robust tracking. Instead, successful systems have to draw from the strengths of multiple cues/modalities. We propose a framework to integrate multiple cues for tracking based on the fusion of dynamic proposal distributions. In the setup of Bayesian sequential estimation, we propose to use the dynamic mixture of proposal distributions to substitute the state prediction probability. The proposal distributions are from different modalities. The fusion problem is then formulated as an optimization problem. The equations to find the global optimal solution is derived based on Karush-Kuhn-Tucker theorem and the approximate optimal solution of the equations is given analytically.

Development and Evaluation of an Alternative ROMP Catalyst for Self-Healing Polymers
Jason Kamphaus (ACS)

Brittle polymers are used extensively in a wide array of industries, including the electronics, aerospace, automotive, and defense industries. The catastrophic failure modes of these materials require careful consideration to their use as well as using time-consuming and expensive non-destructive evaluation techniques for mission critical components. One of these failure modes is microcracking of the matrix. These microcracks coalesce and reach critical lengths that cause failure. A method of repairing these microcracks before they grow and coalesce is to build into the material the ability to react to these microcracks and heal them. The current method for accomplishing this autonomic healing is to incorporate into the material a solid catalyst phase and a microcapsule containing a healing agent. The microcapsules are ruptured by the microcracks, releasing the healing agent into the crack where it reacts with the catalyst phase to form a polymer to bond the two crack faces together. Initial self-healing work has focused on using Grubbs' first generation catalyst in combination with dicyclopentadiene healing agent. Grubbs' catalyst has excellent stability and reactivity but is expensive, has limited availability, and can be deactivated at high temperatures. An alternative catalyst that overcomes some of these limitations is tungsten (VI) chloride (WCl6). Tungsten (VI) chloride's stability and mechanical performance were evaluated to determine its suitability in replacing Grubbs' catalyst in self-healing applications.