Faculty Candidate Presentation

Dimension Reduction Algorithms In Data Mining, With Applications

Jieping Ye
Department of Computer Science & Engineering
University of Minnesota-Twin Cities

Thursday, Mar. 10th
3:30; 380 Dreese Labs
All interested parties are invited.
Refreshments will be served immediately preceding the talk.

Abstract:
Many real-world applications, such as face recognition, text mining, and microarray data classification, involve data of very high dimension. Due to the curse of dimensionality, it is common to pre-process the data and reduce its dimension substantially while preserving essential information.

Traditional algorithms for dimension reduction are based on the vector space model. In this model, each data item is modeled as a vector, and the collection of data is modeled as a single data matrix. In this talk, I will first present the GLRAM algorithm, based on a new data model, where each datum is represented as a matrix. GLRAM computes low rank approximations of a collection of matrices by applying a bilinear transformation on the data. A natural application of this algorithm is in image compression and retrieval, where each image is represented in its native matrix form. Extensive experiments performed using image data show that the proposed algorithm is competitive with traditional ones, such as those based on SVD.

The second part of my talk focuses on some of my other work, including generalized discriminant analysis for microarray data classification and protein structure alignment.

I will conclude the talk with a discussion of my future research plans.

Host: Srinivasan Parthasarathy