Exam schedule, June 14th, Friday.

AM: Digiteo building, room 2014 to be confirmed.

1. 10:00 Luis Fernando Colin
2. 10:30 Shaiful Alam
3. 11:30 Lin Ye

PM: Skype

1. 1:30 Victor Vialle
2. 2:00 Hanna S
3. 2:30 Carla Florencia

Recommendations

The exam involves writing a report on a research paper (2 pages) and presenting a summary thereof orally (slides; 15 minutes + 10 mn questions). The report and the talk must address the following questions:

1. What is the goal
2. Why is it important; what was done before to address this goal, if any
3. What is the idea; what are the key points, what are the difficulties
4. How the idea was validated
5. What is the authors' next step
6. What do you (student) think of it ? Pros and cons.
   1. Remember that among the main things you're supposed to learn during your cursus, one is getting able to appreciate the strengths and limitations of previous works, and plan further original work.

The written reports will be sent to sebag à lri.fr; the presentations will be scheduled according to people availability: if at all possible at the PUIO building; otherwise using skype.

Papers to choose (first come first served)

1. Feature selection, L1 vs. L2 regularization, and rotational invariance
   1. Overfitting is the bad guy of Machine Learning; regularization terms are meant to prevent overfitting. Can such terms achieve feature selection ?
   2. http://ai.stanford.edu/~ang/papers/icml04-l1l2.pdf
2. Error Limiting Reductions between Classification Tasks
   1. How to deal with multi-class discrimination given a binary learner ?
   2. http://www.machinelearning.org/proceedings/icml2005/papers/007_Error_BeygelzimerEtAl.pdf
   3. Lin Ye
3. groupTime: Preference-Based Group Scheduling
   1. Doodle with Machine Learning and negotiation
   2. http://ai.stanford.edu/~ang/papers/chi06-groupscheduling.pdf
   3. Victor Vialle
4. Email Classification with Co-Training
   1. When labeling is expensive, how to make it by using several representations of the data?
   2. https://www.lri.fr/~sebag/Examens/Matwin.pdf
   3. Carla F. Griggio
5. Model Selection via the AUC
   1. Another criterion for ML: The area under the ROC curve
   2. https://www.lri.fr/~sebag/Examens/Rosset.pdf
6. A general method for scaling up ML algorithms; application to clustering
   1. Scaling up by sampling the data: how to do it right ?
   2. http://homes.cs.washington.edu/~pedrod/papers/mlc01.pdf
   3. Hanna Schneider; June 14th
7. Learning with the set-covering machine
   1. Faire des boules d'exemples
   2. http://www.ift.ulaval.ca/~mmarchand/publications/scm-icml.pdf
   3. Shaiful Alam
8. Discovering Communicable Scientific Knowledge from Spatio-Temporal Data
   1. Formuler des lois à partir de données
   2. http://ti.arc.nasa.gov/m/pub-archive/230h/0230%20%28Schwabacher%29.pdf
   3. Luis Fernando Colin