Speaker: Christos Faloutsos,Carnegie Mellon University
Date: January 14, 1999
Indexing and Data Mining in Traditional and Multimedia Databases
The talk presents fast indexing methods for multimedia databases, as well as recent tools for datamining. Specifically, it examines (a) Spatial Access Methods, like R-trees, for multimedia indexing and (b) scaleable methods for lossy compression and rule discovery. For the first part, we do feature extraction, mapping each multimedia object into a low-dimensionality point; then, we store these points in Spatial Access Methods, and thus we can quickly find objects that are similar to a desirable object (e.g., 'find stocks similar to Microsoft'). We describe the conditions under which the method gives no false dismissals, and we also describe FastMap, a method that does automatic feature extraction.
For the second part on data mining, we describe a method that compresses a large data matrix, such as, eg., a matrix with customers as rows, days-of-the-year as columns, and the amount spent in each cell. For such a large, multi-GigaByte matrix, we want to compress it so that (1) it fits on the disk and (2) we can reconstruct arbitrary cells of the matrix quickly. The proposed method exploits patterns in the data matrix, achieves 50:1 compression with less than 10% reconstruction error, and moreover allows visualization.
Christos Faloutsos received the B.Sc. degree in Electrical Engineering (1981) from the National Technical University of Athens, Greece and the M.Sc. and Ph.D. degrees in Computer Science from the University of Toronto, Canada. Dr Faloutsos is currently a faculty member at Carnegie Mellon University. Prior to joining CMU he was on the faculty of the department of Computer Science at University of Maryland, College Park. He has spent sabbaticals at IBM-Almaden and AT&T Bell Labs.
Dr Faloutsos has received the Presidential Young Investigator Award by the National Science Foundation (1989), two ``best paper'' awards (SIGMOD 94, VLDB 97), and three teaching awards.