Possible Mentors

• Georg Holzmann

• Thomas Grill

Description

Support Vector Machines are popular and powerful classifiers/regressors and often used in machine learning problems. In Pd world they could be used for many problems:

• non-linear mapping of sensor data by learning from examples

• audio classification and recognition

• pattern recognition in video data

• gesture recognition, e.g. together with a Wii controller

Currently there is one very popular library for SVMs, called libsvm , which is a C++ library for support vector classification (C-SVC, nu-SVC), regression (epsilon-SVR, nu-SVR) and distribution estimation (one-class SVM). In addition, the same authors created liblinear , which is a linear SVM for large-scale datasets with millions of instances and features.

The goals of this project are:

• write Pd externals which interface libsvm and liblinear

• implement some tutorial style examples, which demonstrate for non-experts how to use these machine learning techniques for (artistic) projects

Resources to start:

• http://iem.kug.ac.at/pd/externals-HOWTO : a tutorial on how to write C externals for pure data

• http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf : a practical guide to SVM classification with libsvm and liblinear

• http://www.csie.ntu.edu.tw/~cjlin/libsvm/ : libsvm code

• http://www.csie.ntu.edu.tw/~cjlin/liblinear/ : liblinear code

Required Skills

• C/C++ programming

• interest in machine learning and support vector machines