TABLE OF CONTENTS

Data Cleaning, despite being a long standing problem, has occupied the center stage again thanks to the mass of uncurated web data and big data. State of the art approaches for data cleaning suffer from two critical shortcomings: they depend on the availability of clean master data (to learn their data generative models), and they assume the feasibility of offline data rectification (so they can use traditional query processing over clean data at run time). To handle these shortcomings, in this paper I propose a novel mediator system called BayesWipe which employs an end-to-end probabilistic frame-work to eliminate dependence on clean master data, and a novel query rewriting model to go beyond off-line rectification to on demand cleaning. I present the details of the framework and demonstrate its effectiveness in supporting query processing over inconsistent data in an unsupervised fashion. i ii ACKNOWLEDGMENTS I thank Dr. Subbarao Kambhampati, who was an excellent advisor to me. — Other acknowledgements here