BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Ph.D. Defense: Fei Xiong
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260916T160621Z
UID:tag:localist.com\,2008:EventInstance_676610
DTSTART:20140407T140000Z
DTEND:20140407T153000Z
DESCRIPTION:Thesis Committee:Octavia I. Camps(Ph.D. Advisor)Mario SznaierDa
 na BrooksJennifer DyThesis Title: Manifold Embedding with Dynamic and/or C
 lassification SupervisionAbstract:It has never been easier to collect data
 . Ubiquitous smart phones and home appliances\, surveillance cameras in pu
 blic spaces\, electronic transactions\, and browsing websites\, are just a
  few examples of ways to generate large complex data that is easily shared
  through the internet. Access to these large datasets brings opportunities
  that range from providing a pleasant experience to an online customer\, t
 o developing smart environments that save energy\, to preventing terrorist
  attacks. Yet\, analyzing and visualizing large\, highdimensional data to 
 enable enhanced decision making and insight discovery remains very challen
 ging. In this thesis\, we propose a set of nonlinear manifold embedding to
 ols that exploit supervised learning information to find low dimensional d
 ata embeddings that preserve spatial and/or temporal correlations characte
 ristics hidden in high dimensional data such as videos and images. The pro
 posed methods extend the maximum variance embedding objective used in the 
 existing Semi-Definite Embedding(SDE) algorithm by incorporating large mar
 gin\, low dynamic order and large margin dynamic classification objectives
 \, respectively. These three different supervision objectives benefit the 
 embeddings with linear separation between classes\, simple dynamics and se
 paration between different dynamics. The proposed algorithms are either fo
 rmulated or relaxed as a convex Semi-Definite Programing(SDP) problem via 
 polynomial optimization and/or low rank matrix approximation technology. T
 he resulting embeddings provide compact\, easier to analyze and visualize 
 representations that capture well the relevant information\, as well as th
 eir relationships\, from the original data. The potential of these tools i
 s illustrated through several challenging applications including classific
 ation\, data visualization\, tracking\, video segmentation and manifold em
 bedding of temporal (eg. videos) to recover missing data and forecast futu
 re measurements.
GEO:42.337532;-71.089317
LOCATION:Egan Research Center\, 206
SUMMARY:Ph.D. Defense: Fei Xiong
URL;VALUE=URI:https://calendar.northeastern.edu/event/phd_defense_fei_xiong
CATEGORIES:PhD Defense
END:VEVENT
END:VCALENDAR