Collaborative/Distributed Machine Intelligence
The research aims to provide intelligence to resource-constraint devices such as surveillance cameras via collaborative inference/learning within a cooperative camera network. The idea at large is to (a) reduce the energy consumption to execute the DNN on the device and (b) improve DNN accuracy by inferencing/learning with other devices in the network.Â
In another project, we explore fusing the multi-modal data streams from heterogeneous devices (UAS and UGV) for improved scene perception.
Indoor Localization for Occupancy Sensing in Smart Building
I lead a research team working on occupancy sensing of a zero energy building in Singapore for (a) predicting HVAC temperature based on building occupancy. We employ RADAR and vision-based Indoor Localization to measure the occupancy real-time as well as store the history for time-series based occupancy prediction to tune the HVAC., (b) employ camera-based luminance (pixel intensity) estimation to estimate the illuminance (LUX) in a given area and subsequently adjust the lights based on the current occupancy and ambient light.
Energy Efficient Multimedia Video Streaming
This project involves setting up Cloud infrastructure, which consists of a data store with videos and Clouldlets that the mobile users access via the WLAN interface. The Cloudlets cache the frequently used videos and live stream for mobile users to access the stream quickly compared to accessing via Cloud.
Energy Efficient Seamless Service Provisioning in Mobile Cloud Computing
I worked on improving the energy efficiency of mobile devices while providing seamless services for mobile users using Cloud Computing features. The thesis work identified offloading computation to the Cloud and performing service level handoff as an optimization problem, which is solved by proposing a mathematical model considering different parameters to calculate risk mitigation before offloading to Cloud.