Vision and Learning Seminar Series

Biweekly on Thursdays at 12pm
VENUE: E305
Coordinators: Prof. Baba C. Vemuri

ARCHIVE OF PAST SEMINARS



Date: Thursday April 8, 2010
Place: Room E305
Time: 12:00 PM
Speaker: Prof. James Hobert
Topic: Improving the Data Augmentation Algorithm


Abstract:
After a brief review of the data augmentation (DA) algorithm, I will introduce a simple modification that results in a new Markov chain that remains reversible with respect to the target distribution. The DA and modified algorithms are compared in the context of a toy Bayesian mixture model where exact eigen-analysis is possible. General results showing that the modified algorithm is always better are presented. I will end by describing a second comparison of the two algorithms in a specific example, this time using a more realistic Bayesian mixture model with normal components.

Date: Thursday March 25, 2010
Place: Room 520A
Time: 12:00 PM
Speaker: Angelos Barmpoutis
Topic: A unified framework for estimating diffusion tensors of any order with symmetric positive-definite constraints
Paper Link: [PDF]


Abstract:
Cartesian tensors of various orders have been employed for either modeling the diffusivity or the orientation distribution function in Diffusion-Weighted MRI datasets. In both cases, the estimated tensors have to be positive-definite since they model positive-valued functions. In this paper we present a novel unified framework for estimating positive-definite tensors of any order, in contrast to the existing methods in literature, which are either order-specific or fail to handle the positive-definite property. The proposed framework employs a homogeneous polynomial parametrization that covers the full space of any order positive-definite tensors and explicitly imposes the positive-definite constraint on the estimated tensors. We show that this parametrization leads to a linear system that is solved using the non-negative least squares technique. The framework is demonstrated using synthetic and real data from an excised rat hippocampus.

Date: Friday March 5, 2010
Place: Room E404
Time: 3:30 PM
Speaker: Prof. Fernando De la Torre
Topic: Learning Components for Human Sensing


Abstract:
Providing computers with the ability to understand human behavior from sensory data (e.g. video, audio, or wearable sensors) is an essential part of many applications that can benefit society such as clinical diagnosis, human computer interaction, and social robotics. A critical element in the design of any behavioral sensing system is to find a good representation of the data for encoding, segmenting, classifying and predicting subtle human behavior. In this talk I will propose several extensions of Component Analysis (CA) techniques (e.g. kernel principal component analysis, support vector machines, and spectral clustering) that are able to learn spatio-temporal representations or components useful in many human sensing tasks. In the first part of the talk I will give an overview of several ongoing projects in the CMU Human Sensing Laboratory, including our current work on depression assessment from video, as well as hot-flash detection from wearable sensors. In the second part of the talk I will show how several extensions of the CA methods outperform state-of-the-art algorithms in problems such as temporal alignment of human behavior, temporal segmentation/clustering of human activities, joint segmentation and classification of human behavior, and facial feature detection in images. The talk will be adaptive, and I will discuss the topics of major interest to the audience.

Biography:
Fernando De la Torre received his B.Sc. degree in Telecommunications (1994), M.Sc. (1996), and Ph. D. (2002) degrees in Electronic Engineering from La Salle School of Engineering in Ramon Llull University, Barcelona, Spain. In 1997 and 2000 he was an Assistant and Associate Professor in the Department of Communications and Signal Theory in Enginyeria La Salle. Since 2005 he has been a Research Assistant Professor in the Robotics Institute at Carnegie Mellon University. Dr. De la Torre's research interests include computer vision and machine learning, in particular face analysis, optimization and component analysis methods, and its applications to human sensing. Dr. De la Torre co-organized the first workshop on component analysis methods for modeling, classification and clustering problems in computer vision in conjunction with CVPR'07, and the workshop on human sensing from video jointly with CVPR'06. He has also given several tutorials at international conferences (ECCV'06, CVPR'06, ICME'07, ICPR'08) on the use and extensions of component analysis methods. Currently he leads the Component Analysis Laboratory (http://ca.cs.cmu.edu) and the Human Sensing Laboratory (http://humansensing.cs.cmu.edu).

Date: Thursday Oct 22, 2009
Place: Room E305
Time: 11:00 AM
Speaker: Karthik
Topic: A Schrodinger Equation for the Fast Computation of Approximate Euclidean Distance Functions
Paper Link: [PDF]


Abstract:
Computational techniques adapted from classical mechanics and used in image analysis run the gamut from Lagrangian action principles to Hamilton- Jacobi field equations: witness the popularity of the fast marching and fast sweeping methods which are essentially fast Hamilton-Jacobi solvers. In sharp contrast, there are very few applications of quantum mechanics inspired computational methods. Given the fact that most of classical mechanics can be obtained as a limiting case of quantum mechanics (as Planck’s constant h tends to zero), this paucity of quantum mechanics inspired methods is surprising. In this work, we derive relationships between nonlinear Hamilton-Jacobi and linear Schrödinger equations for the Euclidean distance function problem (in 1D, 2D and 3D). We then solve the Schrödinger wave equation instead of the corresponding Hamilton- Jacobi equation. We show that the Schrödinger equation has a closed form solution and that this solution can be efficiently computed in O(N logN), N being the number of grid points. The Euclidean distance can then be recovered from the wave function. Since the wave function is computed for a small but non-zero h, the obtained Euclidean distance function is an approximation. We derive analytic bounds for the error of the approximation and experimentally compare the results of our approach with the exact Euclidean distance function on real and synthetic data.

Date: Thursday Oct 15, 2009
Place: Room E305
Time: 11:00 AM
Speaker: Ting Chen
Topic: Group-wise Point-set registration using a novel CDF-based Havrda-Charvát Divergence
Paper Link: [PDF]


Abstract:
This paper presents a novel and robust technique for group-wise registration of point sets with unknown correspondence. We begin by defining a Havrda-Charvát (HC) entropy valid for cumulative distribution functions (CDFs) which we dub the HC Cumulative Residual Entropy (HC-CRE). Based on this definition, we propose a new measure called the CDF-HC divergence which is used to quantify the dis-similarity between CDFs estimated from each point-set in the given population of point sets. This CDF-HC divergence generalizes the CDF based Jensen-Shannon (CDF-JS) divergence introduced earlier in the literature, but is much simpler in implementation and computationally more efficient. A closed-form formula for the analytic gradient of the cost function with respect to the nonrigid registration parameters has been derived, which is conducive for efficient quasi-Newton optimization. Our CDF-HC algorithm is especially useful for unbiased point-set atlas construction and can do so without the need to establish correspondences. Mathematical analysis and experimental results indicate that this CDF-HC registration algorithm outperforms the previous group-wise point-set registration algorithms in terms of efficiency, accuracy and robustness.


Date: Thursday Sep 17, 2009
Place: Room E305
Time: 11:00 AM
Speaker: Guang Cheng
Topic: Non-rigid Registration of High Angular Resolution Diffusion Images Represented by Gaussian Mixture Fields
Paper Link: [PDF]


Abstract:
In this paper, we present a novel algorithm for non-rigidly registering two high angular resolution diffusion weighted MRIs (HARDI), each represented by a Gaussian mixture field (GMF). We model the non- rigid warp by a thin-plate spline and formulate the registration problem as the minimization of the L2 distance between the two given GMFs. The key mathematical contributions of this work are, (i) a closed form ex- pression for the derivatives of this objective function with respect to the parameters of the registration and (ii) a novel and simpler re-orientation scheme based on an extension to the ”Preservation of Principle Direc- tions” technique. We present results of our algorithm’s performance on several synthetic and real HARDI data sets.