Contact

Edo Liberty
Principal Scientist
Amazon (Machine Learing, AWS)
New York, NY
edo@edoliberty.com
עידו ליברטי

Data Mining class description

Data Mining is concerned with efficiently extracting statistics, patterns, structures, or meanings from raw data. This task becomes hard when the amount of data is large, which is often the case in modern data sets.

This year the class will focus mostly on the streaming model. In this model the data is presented to the algorithm one item at a time (items can be: integers, strings, documents, sets, matrix entries, etc...). The algorithm must process the stream under sever space limitations. This is a common scenario in modern mining of massive data sets and presents interesting algorithmic challenges.

We will cover among other topics: sampling, finding frequent items, counting distinct elements, general frequency moment estimation, finding frequent item sets, clustering, dimensionality reduction, support vector machines, online convex optimization, linear regression, and matrix approximation.

I highly recommend that students be familiar with probability theory, combinatorics, linear algebra, basic complexity, and data structures. The class will attempt to be self contained but this is not always possible. Moreover, the class is devoted to ideas, algorithms, and proofs. Students who are interested in explicit data mining applications should not register. To quote a student feedback "you think you're going to get an interesting course and you end up getting a load of math".


Some details...

The class takes place 17:00 to 20:00 every Monday.

Undergraduate students:

Masters and PhD students:

Last year's exams

The following are the Moed Alef and Beit exams from the last two years: Exam Alef 2011 Exam Beit 2011 Exam Alef 2012 Exam Beit 2012

Student projects

You should submit one PDF document containing your project motivation, background, data review, mathematical derivations, experimental results and discussion. You should include any other information you see fit, for example, pseudocode etc.

You are encouraged to use data from Yahoo!'s Webscope. This is a great source for a lot of interesting datasets that Yahoo! occasionally releases. The only thing is that you need to ask for permission to download using your university (.edu) email address. This is simply to identify you as students or researchers. This is strictly procedural and should take not more than a day or so.

The following are a few examples of student projects from previous years. Note that we do not study exactly the same topics every year: project_1 project_2 project_3 project_4 project_5 project_6

The deadline for project submission is May 19th 2013


Discussion Group

http://groups.google.com/group/algorithms-in-data-mining-tau


Class notes and other materials