The following is the schedule of upcoming department seminars. Seminars will be added to the schedule throughout the year. All seminars are free of charge and are open to all members, affiliates, and colleagues of the university community. Seminars are typically held on Tuesday or Thursday at 1:00 pm. If you would like additional information or if you might be interested in presenting a seminar, please contact Helen Cameron or the Department of Computer Science.
Jonathan Boisvert Thesis DefenseWhen: March 03, 2016 @ 12:30pm
Where: EITC E2-528
In seminal contributions to the study of human gaze behaviour, Yarbus examined the characteristic scanpaths that result when an image is viewed, and how these vary depending on the question posed to the observer. Yarbus posited that these scanpaths are sufficiently characteristic that an observer's task might be inferred from observing their gaze patterns. Modern methods in image and data analysis have allowed this hypothesis to be considered directly, in attempting to predict an observers task from their scanpaths. While early efforts of this nature were equivocal, is has since been established that aspects of an observers state of mind or their assigned task may be inferred from observing their gaze.
In this thesis we present analysis for two very different datasets that have not been considered previously in the capacity of predicting task or observer sentiment. One of these involves predicting relatively general tasks assigned to observers before viewing an image, and the other predicting subjective ratings along affective dimensions, recorded after viewing magazine advertisements. The results present a number of interesting observations on the types of tasks or affective categories that can be easily predicted, and the relative value of different measurements for making these predictions. Analysis also demonstrates the importance of how human data is partitioned in predictive analysis, and the complementary nature of gaze specific and image derived features. An important contribution in a broader sense, is the methodology that is proposed in this thesis, which provides strong insight into categories of features most important for different predictions, and the redundancy expressed across different types of features. This is of value for identification of very specific relationships that may benefit from targeted human experimentation (e.g. psychophysics), and also has considerable value in making human centric decisions based on visual media, or in the design of technology that is aware of a user's intent or state of mind.
Dr. Katsuhisa YamanakaWhen: March 08, 2016 @ 1:00pm
Dr. Katsuhisa Yamanaka
Title: Token Swapping Game on Graphs - A Generalization of Japanese Lottery "Amidakuji"
Consider a puzzle consisting of n tokens on an n-vertex graph, where each token has a distinct starting vertex and a distinct target vertex it wants to reach, and the only allowed transformation is to swap the tokens on adjacent vertices. This puzzle is derived by extending Japanese traditional lottery, called "Amidakuji". We investigate how to solve the puzzle with the minimum number of token swaps.