Image Classification using Super-Vector Coding

SpeakerKai Yu
AffiliationNEC Laboratories America, Inc.
Date and Time Sept. 7, 2011, noon - 01:00 p.m.
LocationMcGovern Seminar Room, MIT 46-3189
HostTomaso Poggio

MIT Intelligence Initiative Seminar Series
Speaker: Kai Yu, NEC Labs
Location: MIT 46-3189


Abstract: In this talk I will share our experience about large-scale image recognition using nonlinear encoding of image patches. We worked on extending sparse coding to a broader family of simple coding methods, in particular, super-vector coding, which explore the geometrical structure of sensory image data. The coding of image local features gives rise to significantly better features, which enable simple linear classifiers to produce stronger results, and also scale much better than nonlinear SVMs using Chi-square or intersection kernels. The methods achieved state-of-the-art results on a range of challenging image classification tasks, including Caltech 101, Caltech 256, PASCAL VOC, and ImageNet.


Short bio: Kai Yu is the head of Media Analytics Department at NEC Labs, where he leads a research team working on image recognition, multimedia search, video surveillance, sensor mining, and human-computer interaction. His team has won winner prizes or top positions in various visual recognition evaluations/challenges, including TRECVID, PASCAL VOC, and ImageNet. He co-authored over 70 research papers and has served as area chairs in top machine learning conferences, e.g., ICML and NIPS. In 2011 he taught a class "Introduction to AI" at Stanford University.

There are no comments yet

Post a new comment...