Sahely Bhadra

alt text 

I am a PhD student at the Department of Computer Science and Automation of the Indian Institute of Science , Bangalore, India. I am working in the Machine Learning Lab under the guidance of Prof. Chiranjib Bhattacharyya.

Research Interests

I have been working on building Robust Classifiers to deal with Interval Valued Uncertainty in data as well as kernel matrices. We are also dealing with Scalability and Robustness of these classifiers. In addition to working on the mathematical aspects we focus on applying our formulations on various real life scenarios.

In case of interval-valued data the exact values of the features are not known, whereas statistics like the range of a data point or the region in which the mean lies etc. are known. Such situations are typically common in the domain of medical diagnosis and low level image analysis along with other areas. We have applied our formulation in the field of Micro-array and medical diagnosis dataset.

We have made one of the very first attempts to build classifiers which are robust to uncertainty in the kernel. We have considered various types of uncertainties like Bounded uncertainty and Gaussian uncertainty. This work not only brings in interesting technical challenges to solve but also enables to deal with realistic datasets like protein structures. Hitherto people used to work using the value of the datapoints provided; by absolutely ignoring the resolution information provided. This resolution information actually conveys the fact that the value of the datapoint is uncertain and the amount of uncertainty is related to the resolution information. We have made some formulations based on the Robust Optimization methodology and we have demonstrated in our ICML-10 paper that considering resolution information enbales to deal with protein structures in a more pragmatic way.


IBM Ph.D. Fellowship Award 2010 - 2011.


Sandeepkumar Satpal, Sahely Bhadra, S Sundararajan, Rajeev Rastogi, Prithviraj Sen. Web Information Extraction Using Markov Logic Networks (pdf). 17th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)2011.

Sandeepkumar Satpal, Sahely Bhadra, S Sundararajan, Rajeev Rastogi, Prithviraj Sen. Web Information Extraction Using Markov Logic Networks (Poster). International World Wide Web Conferences (WWW) 2011.

Aharon Ben-Tal, Sahely Bhadra, Chiranjib Bhattacharyya and J. Saketha Nath. Chance constrained uncertain classification via robust optimization (pdf). Mathematical Programming Series B, 2011.

Sahely Bhadra, Sourangshu Bhattacharrya , Chiranjib Bhattacharyya and Aharon Ben-Tal. Robust Formulations for Handling Uncertainty in Kernel Matrices(pdf, demo). International Conference on Machine Learning (ICML) 2010.

Sahely Bhadra, J. Saketha Nath, Aharon Ben-Tal and Chiranjib Bhattacharyya. Interval Data Classification under Partial Information: A Chance-Constraint Approach (pdf). Achieved “Best Runner up” certificate in PAKDD-2009.

S. Bhadra , C. Bhattacharyya , N. Chandra , I.S. Mian. A Linear Programming Approach for Estimating the Structure of a Sparse Linear Genetic Network from Transcript Profiling Data (pdf, demo). Accepted for Journal of Algorithms for Molecular Biology, 2009.

Work Experience

I am working on developing sophisticated machine learning methodologies for modelling the performance of storage system. This is very much usefull to automate storage administration work. Taking information about workload and configuration of storage system as input the model need to predict performance ofthat workload on that storage system. Along with prediction model will also provide confidence about the particular prediction.

I have worked on the problem of extracting structured data from web pages taking into account both the content of individual attributes as well as the structure of pages and sites. We use Graphical Model method to capture both content and structural features in a single framework, and this enables us to perform more accurate inference. The inference problem in our information extraction scenario reduces to solving an instance of the maximum weight subgraph problem. Hence we develop an efficient specialized solver to the inference methods. Experiments with real-life datasets demonstrate the effectiveness of our MLN-based approach compared to existing state-of-the-art extraction methods.


Other Activities