Lifelong Sequential Learning

Project 1:

Melodic Sequence Modeling using Hidden Markov Model

• Board for discussion
• Description
• Overall introduction: Introduction (PDF)
• Supplementary documents
• HMM Tutorial & Explanation of tasks
• Brief tutorial on 'pattern classification'

• S/W Environment
• Matlab
• You can download and install Matlab under SNU campus licence: my.snu.ac.kr → Quick Menu → SW Download → '소프트웨어 목록' → [MathWorks] MATLAB / Simulink
• Tutorials: Matlab Primer by MathWorks, Week 2 materials of Machine Learning class (4190.428)
• If you have problems to get Matlab, you can use GNU Octave as an alternative
• Matlab toolboxes: downloading from the links below is highly recommended
• PMTK3: probabilistic modeling toolkit for Matlab/Octave, version 3. This file contains the whole modules including downloaded modules during initialization. More information on the official website.
• MIDI toolbox: compilation of functions for analyzing and visualizing MIDI files in the Matlab computing environment. This file contains various modifications (by TA) to be used in recent Matlab versions!!. But, if you are interested in the original version, you may visit the official website
• Setting for Project 1 and suppelementary practices
• guide

• Dataset
• For this project, four corpora of MIDI are provided.
• You just need to load 'melody_midi_corpora.mat' into Matlab, which contains just one variable 'midiCorpora' that contains the whole the melodic sequences
• You can see raw MIDI files in 'dataset\raw_midi' folder
• You may check and try process_midi_corpora.m which process whole the raw MIDI files and build  'midiCorpora' variable.

• Practices
• Supplementary Practice 1 (Sep, 10)
• Supplementary Practice 2 (Sep, 12)
• preprocessing codes for MIDI

• References

This page is maintained by Byoung-Hee Kim .
Last update: September 24, 2013.