master's thesis and related research
thesis title: an exploration of feature selection as a tool for optimizing musical genre classification
My thesis concerns the application of optimization techniques (e.g., feature selection) to music classification problems (e.g., artist identification, musical similarity). Questions I'm addressing include: Is feature selection a useful tool for musical classification? What are its limitations? How has feature selection been used in the general machine learning literature? (And what of this literature is useful for music classification?) How should one go about experimentally determining whether an optimization technique is useful in a particular context? I will post some notes and relevant links here periodically.
resources related to my work:
>> thesis proposal
>> my research blog, where I (periodically) keep track of much of my daily work:
http://rfiebrink.blogspot.com/
>> my working bibliography
>> EndNote database for all ISMIR proceedings: ISMIR_EndNote.zip
topics of investigation in spring and summer 2005 (these form much of the background for my thesis):
>> feature selection: comparing algorithms
preliminary testing of outer validation schemes, 16 July 2005
wrapper-based selection and UCI datasets, 20 July 2005
wrapper-based weighting and UCI datasets, 21 July 2005
effect of distance measure on kNN, 28 July 2005
effect of stratification of training and testing sets for kNN, 28 July 2005
population size and mutation rate for JGAP GA feature selection, 8 August 2005
comparing the "ugly blob" on J48 and kNN classifiers, 8 August 2005
>> classification
My
summary table of UCI datasets
>> D2K
NCSA
ALG D2K site
my notes, 18
Jan. 2005
MTCL Presentation,
28 Jan. 2005
notes
on integrating D2K and JGAP, 6 March 2005
notes
on D2K 4.1.1 Basic, 8 March 2005
>> Apple Xgrid
Apple
site
my notes, 1 Feb. 2005
>> Pooch
Pooch
site
>> Grid Weka
Grid
Weka homepage
my notes,
24 Jan. 2005
>> free software for genetic algorithms
my notes,
15 Feb. 2005
notes on JGAP,
1 March 2005