Data Modalities and Differential Privacy
The ubiquity of technology in our modern lives is directly responsible for the explosion in the amount of personal data being collected and analyzed today. One notable feature of this data deluge is that data collection has now extended to different data modalities. Two of the most prominent data modalities are speech and eye-tracking data - recent advances in Automated Speech Recognition (ASR) and mixed reality technologies have led to a steady rise in systems employing speech and eye-tracking as modes of interaction. However, both of these data modalities are rich sources of sensitive information and hence, are vulnerable to privacy threats.
Differential privacy (DP) has emerged as the gold standard for achieving data privacy and has been deployed both by federal and commercial agencies. In this talk, I will show how to extend DP beyond its traditional domain of numeric datasets to address the privacy concerns of other data modalities - specifically, speech and eye-tracking data.
First, I will talk about Prεεch, an end-to-end speech transcription system that (1) protects the users' privacy along the acoustic and textual dimensions; (2) improves the transcription performance relative to offline ASR; and (3) provides the user with control knobs to customize the trade-offs between utility, usability, and privacy.
Next, I will talk about Kalεido, an eye-tracking data processing system that (1) protects the privacy of eye-gaze positions, (2) integrates seamlessly with the existing eye-tracking ecosystems, and (3) operates in real-time.
For both the systems, I will highlight the novelty of our usage of DP that enables the provision of formal privacy guarantee alongside performance that is feasible for real-world applications.
Amrita Roy Chowdhury is a PhD student at the University of Wisconsin-Madison and is advised by Prof. Somesh Jha. She completed her Bachelor of Engineering in Computer Science from the Indian Institute of Engineering Science and Technology, Shibpur where she was awarded the President of India Gold Medal. She has been recognized as a Rising Star at UC Berkeley, 2020. Her research explores the synergy between differential privacy and cryptography through novel algorithms that expose the rich interconnections between the two areas, both in theory and practice.