Orange Forum • View topic - [GSoC 2012] Support for Unsupervised Feature Extraction

[GSoC 2012] Support for Unsupervised Feature Extraction

General discussions about Orange and with Orange connected things (data mining, machine learning, bioinformatics...).

[GSoC 2012] Support for Unsupervised Feature Extraction

Postby freakgeek » Fri Mar 30, 2012 14:29

I would like to work with Orange as part of GSoC 2012(and continue henceforth). Currently I am a Masters students in Mathematics with my bachelors in Computer Science. Over the last year and a half, I have worked on a few ML projects and have a couple of publications(including one at an ACL'11 workshop).

I believe that a widget on Unsupervised feature extraction would come in quite handy by people using Orange. Unsupervised methods have their own beauty and over the last two years, workshops at NIPS[] have shown that unsupervised feature extraction methods outperform the state-of-the-art algorithms without requiring domain knowledge specific to the task at hand.

Last semester I was involved in a research project which involved using dictionary based unsupervised feature learning methods[K-SVD by Elad'07] being applied for Domain Adaptation. Our results were better than the ones already published on the similar datasets. [Abstract:]

Integration with Orange: Output from such a widget could be fed to any other widget for classification, et cetera. The algorithms worth considering include: Dictionary based methods, Sparse Autoencoders among other deep learning architectures.

Such a project would be beneficial both to the research community and the users of Orange(of course, its an overlapping set). Please let me know if this sounds good. I would be really glad to see support for unsupervised learning integrated with Orange.


Return to General Discussions