2011 Poster Sessions : Discriminative Learning of Relaxed Hierarchy for Large-scale Visual Recognition

Student Name : Tianshi Gao
Advisor : Daphne Koller
Research Areas: Artificial Intelligence
Abstract:
In the real visual world, the number of categories a classifier needs to discriminate is on the order of hundreds or thousands. For example, the SUN dataset contains 899 scene categories and ImageNet has 15,589 synsets. Designing a multiclass classifier that is both accurate and fast at test time is an extremely important problem in both machine learning and computer vision communities. To achieve fast evaluation, we exploit the hierarchical structure in the label space by organizing a set of binary classifiers in a hierarchy, where each binary classifier separates two subsets of classes. To achieve high accuracy while being fast, each node in the hierarchy can ignore a subset of confusing classes. We color the classes and learn the induced binary classifier simultaneously using a unified and principled max-margin optimization. We provide a theoretical analysis on generalization error to justify our design. Our method has been tested on both Caltech-256 dataset (object recognition) and the SUN dataset (scene classification), and shows significant improvement over existing methods.

Bio:
Tianshi Gao is a Ph.D. student in the Electrical Engineering Department at Stanford University. He is working with Prof. Daphne Koller in the AI lab of the Computer Science Department to design statistical machine learning algorithms to solve computer vision tasks. He got his Bachelor' Degree with honor from Tsinghua University in 2007 and finished his Master's Degree at Stanford in 2010.