A team of Princeton graduate students in electrical engineering have won a 2019 Innovation Fellowship from Qualcomm Technologies, one of 13 winning teams across the United States to receive mentorship and $100,000 from the mobile technology giant.
Recipients Sameer Wagh and Vikash Sehwag will work to improve security in the branch of artificial intelligence known as deep learning. Demand is soaring for deep learning applications like image processing, speech recognition and language translation. Inspired by the human brain, these applications perform computations over graphs called neural networks to process highly complex information quickly. But sharing vast amounts of data over these networks introduces major privacy concerns for users. Because of the complexity of the networks, engineers have difficulty masking users' information without sacrificing the speeds needed to process so much data in so little time.
Wagh and Sehwag propose adapting an existing security paradigm known as multi-party computation (MPC), which works well on a small scale, for much larger state-of-the-art applications. Their idea introduces a modular approach that could make privacy more efficient at the level of the building block and thus scale up with the size of the network. Their improvements would affect more than 80 percent of current neural network designs, making these MPC security protocols more practical for deep learning technologies.
The proposal was one of 115 submitted for the Qualcomm Innovation Fellowship. Wagh and Sehwag were the only winning team from Princeton. One other team, also working in deep learning, was selected into the first round of consideration for the prize.
The students are advised by Prateek Mittal, associate professor of electrical engineering. Wagh is completing his fifth year as Ph.D. student at Princeton; Sehwag is completing his second year. Both students hold bachelors' degrees from the Indian Institutes of Technology.