Sparse Coding and an Lipschitz Auxiliary Function Algorithm¶
Chris Ding
Professor
University of Texas at Arlington
Abstract :
Sparse coding enforces solutions of pattern recognition methods to contain
mostly zeros and a small number of nonzero elements. This has several
important applications. In feature selection, this enforces entire row of
the regression coefficients to be zero, such eliminates this data/feature
dimension. This L2,1-norm based approach is becoming a popular feature
selection method. In compressed sensing, an input signal (an image or
example) is encoded with s very small number of dictionary signals. This
leads to an improved presentation of the input signal, comparing to
traditional orthogonal basis methods. In this talk, we will briefly explain
these sparse coding methods. In addition, we introduce a new solution
algorithm which uses the auxiliary function approach popularly used in
nonnegative matrix factorization and a Lipschitz continuity condition. This
algorithm can efficiently solve L1, L2,1, L0 norm based sparse coding
problems.
This talk is based on a paper entitled: Towards Structural Sparsity: An
explicit L2/L0 Approach, by Dijun Luo, Chris Ding, Heng Huang. This paper
wons the best-paper-runner-up (2nd best paper) award in ICDM 2010, the
leading international conference on data mining.
Bio:
Dr. Chris Ding earned a Ph.D. from Columbia University. He did research at
California Institute of Technology, Jet Propulsion Laboratory and Lawrence
Berkeley National Laboratory, before joining University of Texas at
Arlington as a tenured professor in 2007. His research areas are machine
mining, bioinformatics, high performance computing, focusing on matrix
tensor approaches. He has served on program committees of NIPS, ICML, KDD,
ICDM, SDM, AAAI conferences. He was funding proposal reviewer for National
Science Foundations of US, Israel, Ireland and Hong Kong. He has given
invited seminars at University of California at Berkeley, Stanford
University, Carnegie Mellon University, University of Waterloo, University
of Alberta, Google Research, IBM Research, Hong Kong University, National
University of Singapore, Beijing University and Tsinghua University. He
published 160 research papers that have been cited 5200 times.