Tinghui Zhou - School of Computer Science - Carnegie Mellon ...

Tinghui Zhou
Email: tinghuiz@cs.cmu.edu
Phone: (612)860-1348
Homepage: http://www.cs.cmu.edu/~tinghuiz/
Education Carnegie Mellon University
M. S. in Robotics (Expected Dec. 2012)
Advisor: Alexei (Alyosha) Efros
GPA: 4.0/4.0
Research
Interests
Research
Experience
University of Minnesota-Twin Cities
B. S. summa cum laude in Computer Science May 2011
Major GPA: 3.95/4.0
Computer Vision: scene and object recognition, image inpainting, image matching, datadriven
image understanding
Machine Learning: max-margin learning, transfer learning, matrix completion/factorization
Robotics Institute, Carnegie Mellon University 09/2011 - Present
Graduate Research Assistant with Alexei (Alyosha) Efros
• Worked on the problem of learning general object models by exploiting the bias across differentimagedatasets.
The proposedframeworkis discriminativeandbased onmax-margin
learning. In particular, our model learns two sets of weights: bias vectors associated with
each individual dataset, and weights that are common to all datasets, which are learned
by undoing the associated bias from each dataset. The common weights are expected to
be our best possible approximation to the object model trained on an unbiased dataset,
thus tend to have good generalization ability.
Disney Research, Pittsburgh 05/2012 - 08/2012
Research Intern with Leonid Sigal
• Worked on the problem of automated photo album creation from unordered image collections.
The problem is difficult as it involves a number of complex perceptual tasks that
facilitate selection and ordering of images to create a compelling visual narrative. To limit
scope, we focus on vacation photographstaken at World Disney World by semi-professional
photographers. To capture and learn rules of album composition, we propose a discriminative
structured model capable of encoding preferences for contextual layout of the scene
(e.g., spatial layout of faces, global scene context, and landmark presence/absence) and ordering
between photos (e.g., exclusion principles or correlations). Once learned, the model
allows automatic composition of photo albums from unordered and untagged collections of
images. (This work has been submitted to CVPR, 2013.)
Computer Science Department, University of Minnesota 09/2010-05/2011
UROP with Arindam Banerjee and Guillermo Sapiro
• Proposed a new matrix completion algorithm, i.e. Kernelized Probabilistic Matrix Factorization
(KPMF), which effectively incorporates external side information into the matrix
factorization process. KMPF explicitly captures the underlying (nonlinear) covariance
structures across rows and columns by working with latent vectors spanning all rows (and
columns) with Gaussian Process (GP) priors. Furthermore, GP priors allow the KPMF
model to fill in a row (or column) that is entirely missing in the original matrix based on
the sideinformationalone, whichis not feasiblefor standardPMF formulation. We demonstrate
the efficacy of KPMF through two different applications: recommender systems and
image restoration.

Center for Distributed Robotics, University of Minnesota 01/2010-05/2011
Undergraduate Research Assistant with Nikolaos Papanikolopoulos
• Worked on a long-term researchproject bringing together domain knowledge from psychiatry,
computer vision, and machine learningto automate the detection ofmental illness with
childhood precursors (e.g. autism). To achieve this, we built a full 3-D monitoring system
with multiple Kinect sensors. The use of multiple views significantly reduces the impact
of occlusion when performing human tracking and motion analysis in cluttered scenes.
• Presented a complete solution to the real-time simultaneous localization and mapping
with a hand-held RGB+Range sensor. At each time step, the sensor gives two types of
measurements: an RGB image taken from a standard camera and an associated depth map
from the range sensor. By utilizing both the color and depth information, the system is
able to localize itself with respect to a predefined world coordinate frame and reconstruct
a realistic 3D model of the environment.
• Developed an algorithm for efficient detection of mutually-occluded ellipses. Randomized
Hough transform (RHT) is known for its computational efficiency in ellipse detection.
However, it performs poorly if the target ellipses are mutually-occluded. We developed an
RHT extension that utilizes K-means clustering to improve its accuracy and efficiency in
detecting mutually-occluded ellipses.
Publications A. Khosla, T. Zhou, T. Malisiewicz, A. A. Efros, and Antonio Torralba, “Undoing the Damage
of Dataset Bias,” European Conference in Computer Vision (ECCV), 2012.
Leadership
Activities
Programming
Skills
Selected
Awards
T. Zhou, H. Shan, A. Banerjee, and G. Sapiro, “Kernelized Probabilistic Matrix Factorization:
Exploiting Graphs and Side Information,”SIAM International Conference on Data Mining
(SDM), 2012.
T. Zhou and N. Papanikolopoulos, “Enhancing Randomized Hough Transform with K-Means
Clustering to Detect Mutually-Occluded Ellipses,” Proceedings of the 19th Mediterranean Conference
on Control and Automation (MED), 2011.
Board Member of A-PACE 09/2012-Present
• Co-organized discussion panels and invited talks for the Association of Pittsburgh Area
Chinese Entrepreneurs (A-PACE), which seeks to provide a platform for Chinese students
in Pittsburgh, especially those from CMU and University of Pittsburgh, to exchange entrepreneurial
ideas and establish potential collaborations.
Coordinator of Technology Day Camp 08/2010
• Served as a coordinator of the annual Technology Day Camp hosted by the Center for Distributed
Robotics at UMN. The mission of this event is to introduce the latest technology
development to underprivileged youth in the Minneapolis/St. Paul area.
Languages: C/C++, Matlab, Lisp, HTML, MIPS assembly
Libraries and Tools: OpenGL, OpenCV, L ATEX, LIBSVM, Aria, Player/Stage,MobileSim, GDB,
Subversion, Mobile Robot Programming Toolkit.
UROP Award University of Minnesota 2011
IT Merit Scholarship University of Minnesota 2010
IMA MCM Meritorious Winner University of Minnesota 2010
Summer Undergraduate Research Fellowship Purdue University 2010
University Honors Program University of Minnesota 2009