Presentations
The presentations are the second most significant part of your grade
in this class (20%), and provide an additional way to gain a deeper
understanding of a specific topic.
Please come and see me as early as possible to pick-up a paper.
Prepare three (3) suggestions for papers you would like to present.
Due
Dates
- Paper selection: Tuesday, December 12. Email
me.
- Slides: Two days before your talk. Meet
with instructor.
- Presentation: In Class.
- Sunday, January 3.
- Sunday, January 10.
Guidelines
There are three themes which extends the material we
learn in class. It can be a short conference paper or a long journal
paper. Each presentation will be twenty minutes long (20) with an
additional five minutes (5) for questions and comments. You should
decide what part or the paper you want to present. (For example the
details of an algorithm? analysis? specific set of experiments?). You
should use slides for the presentation (your favorite application). At
least three days before your presentation (Thursday, the week before) you should meet with me to discuss the slides and presentation.
You can choose
to present either a conference paper or a journal paper. UAI (http://www.auai.org/) , NIPS (http://nips.cc/) and ICML are three of the
main computer science conferences where graphical models work is
published, with many of the recent papers online. Of course,
conferences on computer vision, natural language processing,
computational biology, sensor networks, and many others have a lot of
graphical model papers as well. JMLR (http://www.jmlr.org/) is a main
relevant online journal, but there are many more, printed or
electronic, available through the library.
Themes
There are three themes. Some of them overlap with the material
taught in class and provide a way to gain a deeper understanding in
a specific subject; some are about new material
- Training: Fast and/or specialized methods for learning the paramters of a model.
- Semisupervised and Unsupervised Learning: Learning with indirect feedback in the context of graphical model is challenging and an open field of research.
- Structured Prediction: Learning models for making
multiple, related predictions at once. Representative examples include
pixel-labeling or parsing a sentence, where the
input is a sentence, and the output is a parse
tree of a sentence.
Evaluation
We will evaluate your presentations according to the following four criteria:
- Content: Is a good range of information included (not too obvious nor too specialized)? Are the main points covered?
- Delivery: Is the presentation clear and concise?
Well practiced?
- Organization: Is the talk well-structured? Easy to follow?
Some example papers:
-
-
Training:
-
Unsupervised and Semisupervised Learning:
-
Structured Prediction: