sahely

Research interest:

Machine Learning (learning from noisy and incomplete data), optimization (robust optimization and convex optimization for large data), data analysis, network analysis, bio-informatics (micro-array data analysis and molecular network analysis) .

Summary:

I am currently working on completing incomplete graph or network using machine learning models and on handling missing data for kernel method when data has multiple sources. Previously, I have worked on robust optimization to learn form noisy data. My recent work proposes a new pre-processing method to correct noisy labels via mutual consistency check using a Parzen window classifier. It has also used the Spannogram framework to solve the problem with approximation guarantees. During my graduation I have been working on building robust classifiers to deal with interval valued uncertainty in data as well as kernel matrices. I have also dealt with scalability and robustness of these classifiers. A novel distribution free large deviation inequality has been proposed which handles uncertainty. A mirror descent algorithm (MDA) like procedures have been applied to make learning of proposed robust formulations scalable. In addition to working on the mathematical aspects I focused on applying new formulations on various real life scenarios.

Awards:

• IMPECS postdoc fellowship, 2012-2014.
• IBM Ph.D. Fellowship Award 2010 - 2011.
• Best Runner Up paper award in PAKDD, 2009.