Hamm, Martinez receive Google Faculty Research Awards
Their projects were among the 113 selected for funding out of the 805 proposals submitted worldwide.
Electrical and Computer Engineering Professor Aleix Martinez received $46,361 in support of his online tutor project, which aims to make American Sign Language (ASL) more accessible by providing critically needed feedback to students.
The problem his research aims to solve, Martinez said, is that existing classroom and educational materials provide little feedback for students trying to learn ASL.
Through the use of a video camera connected to a personal computer, the ASL machine vision software created within Ohio State’s Department of Electrical and Computer Engineering, not only demonstrates signing, but also automatically analyzes the user’s imitation. The software picks up on hand shapes, movement types, and places of articulation or even facial expression using machine vision algorithms and a webcam or built-in computer camera. Non-rigid structures from motion algorithms then help reconstruct the face, torso and hands of the student in 3-D. A graph model then derives a metric for classification of the recovered hand shape, motion and place of articulation, providing useful feedback to the user.
Jihun Hamm, a research scientist and lecturer in computer science and engineering, received $45,000 for his research that will allow smart devices to share data and train learning algorithms, while preserving users’ privacy.
Effective privacy mechanisms for smart devices are still underdeveloped, said Hamm. Using machine learning techniques, information from smart devices can better serve individuals’ needs by predicting user behavior and automating daily routines. But this also raises privacy issues for users, as sensory data are usually personal and may contain sensitive information.
Hamm’s work presents new directions in private learning with smart devices using two complementary approaches: local differentiallyprivate algorithms for learning with smart devices and feature transformation for anonymous learning.
Differential privacy aims to provide the means to maximize the accuracy of queries from statistical databases, while minimizing the chances of identifying its records. Basic mechanisms of differential privacy are perturbation and random noise addition, which allow third parties to learn useful knowledge from population data, but prevents them from learning about individual data accurately.
Complementary to this approach, Hamm will also use feature transformations to fill the gaps of differential privacy, so that a noninvertible transformation of private data can be learned from data to achieve an optimal tradeoff between data utility and user anonymity for a large class of learning algorithms.
Google’s Faculty Research Awards program supports world-class faculty pursuing research by covering one year of tuition and expenses for a graduate student, and provides both faculty and students the opportunity to work directly with Google researchers and engineers.
