Overview
The MDL approach began with Kolmogorov's theory of algorithmic
complexity, matured in the literature on information theory,
and has recently received renewed interest within the statistics
community. By viewing statistical modeling as a means of generating
descriptions of observed data, the MDL framework (link to Barron,
Rissannen and Yu, 1998, and Hansen and Yu, 2001) discriminates
between competing model classes based on the complexity of each
description based on a model class. Precisely, the Minimum Description
Length (MDL) Principle states that
Choose the model that gives the shortest
description of data.
The complexity of a description is measured by the code length
for the data based on the model.
Yu's group's another interest has been Audio Compression. Recently
Gerald Schuller, Bin Yu, Dawei Huang, and Bern Edler developed
a coding method which achieves leading compression ratios and
a low lag for a wide variety of audio sources. Bin's group's
work was focused on using prediction method to reduce redundancy
which has strong connection to boosting, competitive on-line
statistics and MDL.
People to Contact
Peng Zhao,
Bin
Yu
Related Publications
- Mark Hansen and Bin Yu (2002). Minimum Description
Length Model Selection Criteria for Generalized Linear Models.
Tech. Report 619, Statistics Dept, UC Berkeley.
- Mark Hansen and Bin Yu (2000). Wavelet thresholding
via MDL for natural images. IEEE Trans. Inform. Theory
(Special Issue on Information Theoretic Imaging). vol.
46, 1778-1788.
- R. Jornsten, W. Wang, B. Yu, and K. Ramchandran (2002).
Microarray image compression:
SLOCO and the effects of information loss. Signal
Processing Journal (Special Issue on Genomic Signal Processing).
(accepted). Tech. Report 620, Statistics Dept, UC Berkeley.
- Gerald Schuller, Bin Yu, Dawei Huang, and Bern Edler (2002).
Perceptual Audio Coding
using Pre- and Poster- Filters and Lossless Compression. IEEE
Trans. Speech and Audio Processing. Vol. 10 (6), 379-390
more...
|