final project

This course includes a final project. 1/4 Of your final grade will be computed from your final project grade.

The final project should allow you to work more in depth on a specific aspect of machine learning that you are interested in. In general, you can come up with any machine learning related idea as your project. The most straightforward ideas probably revolve around applying a machine learning technique to several interesting data sets. Or maybe you have an idea for improving a specific algorithm? Or want to analyse some theoretical aspect in more detail? Maybe you are more into visualisation and have an interesting idea how to connect that to Machine Learning, that's fine, too.

Some examples:

• if you found some interesting data that you want to analyse and you are looking for  a suitable algorithm, check out recent publications in the community (from about 2005 on, see NIPS, ICML, AISTATS, CVPR,  UAI, ...)
• a specific, but unusual idea: Implement neural networks in Lua, usingLuaJIT FFI to a BLAS (preferable GotoBLAS). Compare to a C/C++ based implementation with respect to execution speed
• Analyse tweets using the recent paper Learning Continuous Phrase Representations and Syntactic Parsing with Recursive Neural Networks
• Implement an SVM on a GPU using cudamat (look at PEGASOS subgradient method, and/or O. Chapelle's 2007 paper, not sure if this is feasible, should be so for linear SVM)
• use a differentiable unsupervised preprocessing technique, combining it with any supervised learning approach that can be trained with gradient descent.
• LeadLag LDA: Estimating Topic Specific Leads and Lags of Information Outlets
• TopicFlow model: Unsupervised learning of topic specific influences of hyperlinked documents.
• For NLP tasks, SENNA (http://ronan.collobert.com/senna/) could be very helpful.

but don't bother us with an implementation of backprop in C (which you can download from several thousands sources on the web)!

Proposals for final projects must be handed in by email to smagttumde by December 16, 2011, 23:59:59.00 CET. We will decide on your proposal within the week after that and advise you to follow your idea, make some changes, or def it. You can decide to do your final project by yourself or in a group of 2.

You have to hand in your final project before March 16, 2012 CET. Apart from a written summary (3-6 pages) and a poster (both to be handed in in PDF format only), you have to submit all used code and used datasets. You are free to give a short presentation (about 10-15 minutes)---by appointment only.

Looking for some inspiration? Check out the final projects of Andrew Ng's course at Stanford: http://www.stanford.edu/class/cs229/projects2010.htm.

If you are looking for data sets, take a look at archive.ics.uci.edu/ml/. A bunch of machine learning software can be found at mloss.org/software/.