Pooja Jhunjhunwala
Pooja Jhunjhunwala is a Staff Software Engineer and Tech Lead at Google, working on Pixel Biometrics. Her team works on the ML Systems that powers the industry’s first Class 3 Biometric Facial Authentication system using standard RGB cameras, a breakthrough that brought financial-grade security to millions of Pixel devices without specialized hardware. With over a decade of industry experience and a Master’s in Computer Science from Texas A&M University, she has also led critical on-device ML projects for Google Clips, Intelligent Photography, and Pixel 4’s NIR Face Auth system. Her work focuses on democratizing high-security biometric authentication and advancing privacy-preserving machine learning at the edge.
Authored Publications
Sort By
Preview abstract
Attorneys, judges, and others in the justice system are constantly surrounded by large amounts of legal text, which can be difficult to manage across many cases. We present CaseSummarizer, a tool for automated text summarization of legal documents which uses standard summary methods based on word frequency augmented with additional domain specific knowledge. Summaries are then provided through an informative interface with abbreviations, significance heat maps, and other flexible controls. It is evaluated using ROUGE and human scoring against several other summarization systems, including summary text and feedback provided by domain experts.
View details
Preview abstract
Raspberry Pi is revolutionizing the computing in-dustry. Originally designed to provide low cost computers to schools it quickly expanded far beyond that. With this inexpensive technology you can accomplish tasks previously unexplored. One such set up is a cluster computer to run parallel jobs. Many systems built for parallel computing jobs are either very expensive or unavailable to those outside of academia. Supercomputers are extremely expensive to own, use, power and maintain. Even though average desktop computers have come down in price the expense can still get quite high if you need a larger amount of processing power. In this project, we take ten Raspberry Pi 2 computers and connect them over an ethernet network to build a parallel version of a supercomputer. We show how one can be built inexpensively and compare the performance of our system to both desktop machines and supercomputers in use by academia. We look at a price comparison as well as performance and show the differences between the older versions of the Raspberry Pi cluster computers and our own implementation. Our results show that our Pi cluster implementation significantly outperforms previous Pi cluster projects with the cost being similar. This project strongly made us believe that the Raspberry Pi is a great learning platform for academia and personal use and the processing power at the cost involved is hard to beat.
View details
