Paper presentation

For the paper presentation, you should choose one of the papers below or another paper that deals with machine learning that has been published in a high-quality conference or journal such as ICML, JMLR, Machine Learning, AAAI, etc. Once you choose the paper you should email me with your choice.

Once you have chosen a paper, you should create a 20-minute presentation on that paper, which you will present to the class. You should use overheads such as PowerPoint or similar. In order to present the paper, you must understand the paper, so you should read it very carefully and understand it. Your job as the presenter is to pull out the important points of the paper and present them clearly for the audience. If some part of the paper is unclear, your job is to make it clear.

Here are a list of questions you should answer for yourself and your audience when you read the paper and when you give your presentation.

What is the main point of the paper?
What are the authors doing that is new?
What is the task that the authors are trying to accomplish?
What models and learning algorithms are used in the research?
How are the author's claims being evaluated -- using experiments, theoretically, other ways?
What are strong points of the paper?
What are weak points of the paper?

You should not try to answer these questions directly in the presentation (i.e. you should not say "The main point of the paper is... The authors new contribution is..."). Instead, you should highlight these things as you present the paper.

Presentation schedule

There will be at most three presenters per class period.

April 7th: Frank, Firasath
April 12th: Nick, Nerissa

Presentation requirements

Here is what is required for this project:

Choose a paper and have it approved by the instructor.
Sign up for a slot to speak on April 7th or 12th.
Read the paper carefully (start immediately on this).
Meet with the instructor individually to discuss the paper (starting Wednesdsay, March 30th).
Develop slides to present the paper.
Meet with the instructor to go over the slides prior to the presentation (on April 5th and 6th).
Send your completed slides to the instructor on the day before your talk.
Give your talk on either April 7th or 12th (2 or 3 talks on each day).

Some guidelines

Here are some notes on giving a research talk, from Charles Elkan.

Here are some additional points, from your instructor.

Make sure you put some thought into choosing your paper, to minimize frustration later.
Choose a paper that looks interesting to you.
Present the research in the paper. If you talk about your opinion or something outside of what the paper claims, make sure you make that clear.
Use graphs that are in the paper.
Explain things for your audience. What might seem obvious to you after reading the paper will not be obvious for your audience.
Have fun with the presentation!

Evaluation

You will be evaluated according to the guidelines listed here, and will receive feedback by this feedback form.

Paper list

These papers have been chosen as being interesting and related to the topics we have covered in this course. Since research papers are novel by definition, most of these will deal with something you are not yet familiar with. Use this as an opportunity to learn about the topic, and ask your instructor questions if you are confused about something.

Nerissa -- Beyond independence: conditions for the optimality of the simple bayesian classifier. Pedro Domingos and Michael Pazzani, In proceedings of the thirteenth International Conference on Machine Learning, 1996.
A brief introduction to boosting. Robert E. Schapire. In proceedings of the sixteenth International Joint Conference on Artificial Intelligence, 1999.
An evaluation of machine-learning methods for predicting pneumonia mortality Gregory F. Cooper et al. Artificial Intelligence in Medicine volume 9, 1997.
Frank -- Learning to select good title words: an new approach based on reverse information retrieval. Rong Jin and Alexander G. Hauptmann. In proceedings of the eighteenth International Conference on Machine Learning, 2001.
Obtaining calibrated probability estimates from decision trees and naive bayesian classifiers. Bianca Zadrozny and Charles Elkan. In proceedings of the eighteenth International Conference on Machine Learning, 2001.
Nick -- Feature selection for high-dimensional genomic microarray data. Eric P. Xing, Michael I. Jordan, Richard M. Karp. In proceedings of the eighteenth International Conference on Machine Learning, 2001.
Efficient exact k-nn and nonparametric classification in high dimensions. Ting Liu, Andrew W. Moore, Alexander Gray. In proceedings of Neural Information Processing Systems, 2003.
Firasath -- A comparative study on feature selection in text categorization. Yiming Yang and Jan O. Pedersen. In proceedings of the fourteenth International Conference on Machine Learning, 1997.

valid html and css