The fellowship mainly investigates an analysis of the state-of-the-art approaches, design and implementation of cutting-edge deep neural network models to be used on a mobile platform. It explored ways to optimize the deployment of these machine-learning models for prediction tasks on the mobile devices which requires energy efficiency and accuracy.
Optimizing a program for Graphics Processing Units (GPUs) is critical for performance, yet remains a challenge due to the non-intuitive interactions among the optimizations and the GPU architecture. Automatic optimization tuning for a GPU is demanding particularly given the exploding number of mobile GPU variants in the market.
Driver distraction has long been a critical issue drawing substantial amount of research effort. In order to reduce driver distraction for improved driving performance and safety, automotive suppliers have been endeavoring to provide optimum user interaction solutions. Until recent years, there have been growing interests in the use of gestural interfaces for in-vehicle information systems; however, little is known about how such gesture-based interactions differ from existing touch- and voice-based interactions in the context of driver distraction.
More than 25% of the fatal and injury car crashes are related to fatigue or drowsiness. This calls for the need for designing automated driver monitoring systems, which can continuously measure the drivers' vigilance level and alert them if their cognitive state is not safe for driving anymore. One of the most reliable solutions is to directly measure the electrical activities of the brain to monitor the driver's cognitive state. The proposed research aims to design a non-intrusive yet efficient monitoring system that uses electroencephalogram (EEG) signals.
The goal of this project is to design a monitoring system which continuously observes the driver's cognitive state and provides a realtime feedback for the driver regarding his/her mental state and vigilance level, as described in the “Background Information” section. Studies in [5-9] have examined the changes in the theta, alpha, and beta EEG rhythms during different driving tasks and have shown a significant correlation between the changes in spectral power of EEG signals and the driver's workload and fatigue level.
Conventional camera sensors record three color channels: red, green and blue. In this project we will investigate computational photography algorithms for cameras that record a near-infrared channel (NIR) in addition to RGB. This channel is particularly useful for biometric imaging and holds great potential in consumer imaging applications as well. The key challenge in simultaneously capturing RGB and NIR is that lens behavior depends on wavelenth and thus the NIR channel may be defocused compared to the other three.