The following presentations will play the role of the midsem exam
in this course. They will carry 50% weight (the remaining 50%
will come from the semstral exam). A typical presentation will be
45 min in duration (more if Rizu and Noirrit wants a joint
presentation). Remember the following points
regarding the presentations:
Presentationn skill is one of the most important skills that
a human being can possess. This skill, like many other skills,
can be perfected only through practice. However, it is not often
possible to get a set up for practising this skill. The
presentations will give you a chance to improve this skill. Treat
this as an opportunity to experiment, and not just a boring
compulsion.
Focus more on how you present, than on some
mathematical detail.
Above all, try to be imaginative in your
presentations. Make it as entertaining as possible (without
sacrificing the content, of course!). Using a snazzy slide
transition/sound effect might be cool, but giving a new example
is better.
Don't panic about goofing up! Nobody learns without goofing
up someime or other. If you have tried sincerely then your
grades will definitely reflect that!
Topics
Recognition of printed digits using PCA:
Who: Adhideb Biswas
What: Use the data given here to see how
well PCA performs for printed digit recognition.
When:
Inside CART:
Who: Tamalika Koley
What: The presentation should focus on proving the
optimality of CART, and also talk about how missing values are
handled in CART. The main reference will be the CART book.
When:
Linear and polynomial separability in pattern recognition:
Who: Jayant Jha
What: The main aim is to
present this paper. The focus
should be on rigorous presentation of at least the first three pages.
When:
Recognition of handwritten letters by SVM
Who: Rizu Mukherjee and Noirrit Kiran Chandra
What: Download a data set of handwritten English lower
case letters from this link. Your job is to
build a recogniser using SVM. Feel free to experiment with
different kernels. You may also use other techniques with a view
towards comparing them with SVM.
When:
Kernel PCA
Who: Debolina Ghatak
What: Kernels allow linear methods to be applied
to nonlinear set ups. Little wonder that we have a kernel version
of PCA. The aim here is to explore this idea. You may
use this paper as a starting point.