Postdoctoral Fellow, Applied Mathematics
Thanks to the Herman Goldstine Postdoctoral Fellowship in Mathematical Sciences, I currently have a one year fellowship in the Business Analytics and Mathematical Sciences department at IBM Research, Yorktown Heights. From 2011 to 2013 I had a two-year fellowship from the Fondation Sciences Mathématiques de Paris and worked at Paris 6 (JLL lab) with Patrick Combettes. Until 2011, I was at Caltech supervised by Emmanuel Candès.
Office telephone: (914) 945 2074
I am interested in collaborating with mathematicians, scientists and engineers on diverse projects.
Optimization: first-order methods, quasi-Newton methods, primal-dual algorithms
Mathematical applications: compressed sensing and variants, matrix completion and variants (robust PCA…), non-negative matrix factorization and end-member detection, sparse SVM
I will be joining IBM beginning in Sept 2013 as a Goldstine postdoc
This website has been redone using jemdoc along with some customizations to the python script. Jemdoc allows content pages to be very simple, and makes using integrating equations into webpages very easy.
I've made a little website for our A2I project.
Matlab/Octave code for the 0-SR1 method mentioned in A quasi-Newton proximal splitting method should be available very soon! Please check back.
Some tips on using Mendeley. I have been using Mendeley to organize my papers for about 3 years now, and I want to share two quick tips that make it work even better. The first is that there is preliminary support in their native PDF viewer to use PDF bookmarks (the kind that work well in Acrobat). To enable this feature, run Mendeley like mendeleydesktop –pdf-toc (for linux; for Windows, it is possible as well but a different command). The second feature is using a more advanced search, and I don't need to describe it because it is covered well in this article on Mendeley search.
My preprint A quasi-Newton proximal splitting method with Jalal Fadili is now available. Accepted in NIPS 2012.
At SIAM IS ’12 in Philadelphia, I describe a new quasi-Newton method for non-smooth or constrained problems (see the preprint with Jalal Fadili). Slides
Just released TFOCS v.1.1, which is a major update of the old TFOCS release.
The sparse- and low-rank solver wiki is now online! This replaces the old list of algorithms that I had. The whole world is invited to join and edit the wiki. The benefits of the wiki format are numerous: it can stay up-to-date, it allows anyone in the world to contribute (including people who write algorithms, people who use algorithms, people who are doing background research, etc.).
Final version of my thesis, Practical compressed sensing : modern data acquisition and signal processing, is online.
“TFOCS: Flexible First-order Methods for Rank Minimization”, at the Low-rank Matrix Optimization minisymposium at the 2011 SIAM Conference on Optimization in Darmstadt. Here's the RPCA video on youtube (Matlab code to generate this is available at the RPCA demo page).
“Templates for Convex Cone Problems with Applications to Sparse Signal Recovery” by S. Becker, E. Candès, M. Grant is available. PDF, Software
I am the TA for ACM 114, Parallel Algorithms for Scientific Applications.
“Quantum state tomography via compressed sensing” by Gross, Liu, Flammia, Becker and Eisert, is on the arXiv. Link
I just started using Mendeley to organize my PDFs. So far, it's exactly what I had been looking for and I highly recommend it. Several other students in my department started using it as well. On a similar note, many people have recommended Dropbox for file storage and syncing, and some collaborators are using it for editing a paper. Edit (2012): I now prefer spideroak but still use Dropbox for legacy reasons. I also tried ubuntu one for a year and was unhappy with it (at least with Ubuntu 10.04).
svt.stanford.edu is a software package for Singular Value Thresholding, for Matrix Completion problems. Please email me if you have questions about the software (and include information on the version of Matlab, the operating system, and the hardware – e.g. 32-bit or 64-bit). We are working on making it more compatible with 64-bit systems and compatible with complex-valued data.