I am a PhD student with Computer Science department at IIT Bombay advised by Prof. Sunita Sarawagi and Prof. Soumen Chakrabarti. Before joining IIT Bombay, I had spent wonderful two years as a research member staff at Amuse Labs contributing features to a digital archival project called ePADD. I graduated with computer science majors from IIT Mandi where it all started :). I work in the area of Applied Machine Learning with emphasis on problems arising due to domain shifts. See Research Theme for an elaborate version.
Research Theme
Learning systems pick on spurious correlations in the dataset and fail to generalize. Thereby, the predictions during deployment are counter-intuitive and are not foolproof. I plan to investigate dataset biases that enable such overfitting and design algorithms for generalization. Contemporary learning systems when they pick on non-causal features resemble Cargo Cult Science. Mechanisms that enable predictions with only the right causal features will bring us closer to building machines with human intelligence.
Humans are equally adept at fast adaptation as they are at quick generalization. For example, it does not take a person long to drive a Nissan car after having driven a Ford car. Efficient adaptation is crucial for building systems tailored for personal or organizational use. Adaptation is indispensable at catering to the population that were under-represented during the train time.