On February 3, the Women in Machine Learning (WiML) organization profiled CSE Prof. Kamalika Chaudhuri on its Facebook page. The Feb. 3 article notes that her research is on "the theoretical foundations of machine learning, and she works on designing machine learning [ML] algorithms with rigorous performance guarantees."
Chaudhuri (at right) is particularly interested in privacy-preserving machine learning -- how to design ML algorithms that operate on the sensitive data of individuals, while ensuring that their outputs are rigorously guaranteed to preserve privacy. "An obvious approach to privacy is anonymization -- removing names and addresses and the obvious identifiers from the data, and then using the rest of the data for learning," said Chaudhuri. "These approaches however are insufficient for privacy. Even without obvious identifiers, personal data tends to be very unique, and it is often possible to automatically link together different sources of information and re-identify individuals in anonymized data."
The computer scientist and former CSE postdoctoral researcher (2007 to 2010) went on to explain her focus on guaranteeing 'differential privacy', a rigorous definition of privacy designed by cryptographers in 2006. "Differential privacy is typically obtained by randomly perturbing the result of a function -- which could be as complex as a classifier or a clustering -- computed on the sensitive data," said Chaudhuri. "The challenge is to design algorithms that can achieve privacy as well as high accuracy given a certain number of samples."
Chaudhuri is also studying privacy challenges that arise in more correlated and more complex data, such as time series and social networks, and she has been working on "algorithms that can compute statistics on such data while still guaranteeing a generalization of differential privacy."
Learn more about Women in Machine Learning at the WiML Facebook page.