NU ML Journal Club first meeting!
Details
Join the NU Machine Learning Meetup for its first Machine Learning Journal Club meeting!
Journal Club meetings will each focus on an advanced topic in the machine learning field through study and discussion of a high impact journal article (or related articles). One member of the attendees will give a brief overview of the article’s contents and will lead group discussion. We will meet in the smaller EECS conference room Ford 3.340 – a great venue with lots of whiteboard space. We will hold regular Journal Club meetings the last Thursday of the month – and the first will be June 27 from 5pm – 6pm.
All members are encouraged to suggest papers/ideas for Journal Club meetings!
I’ll start the ball and for our first meeting. We will discuss the Subspace Clustering model as well as a few algorithms using the articles
“A Tutorial on Subspace Clustering” by Rene Vidal
http://cis.jhu.edu/~rvidal/publications/SPM-Tutorial-Final.pdf
“Sparse Subspace Clustering” by
http://vision.jhu.edu/assets/SSC-CVPR09-Ehsan.pdf
Subspace Clustering is a generalization of standard dimension reduction where data is assumed to approximately lie on a single lower-dimensional subspace (common techniques like PCA are then used to recover this subspace). The assumption in Subspace Clustering is that the data lies on a union of lower dimensional subspaces , not just a single subspace (see the nice illustration of the idea on the first page of the first paper above).
Subspace Clustering is a hot technique that has been applied to a wide array of applications in machine learning, image processing, large scale data mining, prediction of dieseases, recommendation systems, as well as many others (a quick web-search for “Subspace Clustering” and any one of these will return an appropriate application paper). Our first paper is a general tutorial on an array of classic algorithms for the model, and the second is an introduction to a cool new algorithm for Subspace Clustering that leverages sparsity.
PS: There are a ton of algorithms for attacking this problem (the tutorial describes many of them in detail). In my review/intro I will focus in on describing the Sparse Subspace Clustering method described in the second paper above. This algorithm (along with several others described in the tutorial paper) rely on spectral clustering - a common clustering technique. Here's a tutorial on spectral clustering for those interested