Projects Involving Lab Members
Real-time Video Analytics
Publications at USENIX ATC 2024, ACM TOMM 2025, COMSNETS 2024, MobiCom 2024 Workshop, COMSNETS 2025 Poster, CCS 2025 Workshop (Funded by Cisco University Research Fund and TiH iHub-Anubhuti)
Modern cities and transportation systems utilize data from thousands of cameras, usually analyzed using deep neural networks. Drawing intelligence from this data, however, requires sending this data overall a wireless network, proper data ingestion and minimization of unnecessary computation. This project identifies ways of filtering data using lightweight techniques on the camera-side, as well as techniques of reducing computation on the server-side. Many of the techniques developed in this project have been prototyped and are now open-sourced to the community.
Tail Latency for Edge Computing
Publications at COMSNETS Poster 2024, IEEE TNSM 2025 (Funded by ANRF and DST)
Many IoT applications used today not only require real-time results, but also require such real-time results with high reliability. Such high reliability typically requires even high percentile latencies of 90\% or even 99\% to be below a certain limit. In this project, we investigate ways of scheduling tasks on edge devices such that such high percentile tail latencies satisfy such requirement. We have specifically looked at reinforcement learning that balances the use of redundant execution with reduction of tail latencies for different tasks.
Cloud-Assisted Autonomous and Safe Driving (Funded by DST-PURSE Project)
Publications at COMSNETS Poster 2025
Although autonomous vehicles (AVs) are being trialed, they are very expensive currently due to the requirement of incorporating GPUs with large memories. This project studies techniques of offloading some of this computation to cloud/edge servers. These techniques include adjusting the quality of videos depending on the weather, re-training the models when the accuracy drops as well as identifying the tradeoffs in terms of speed and acceleration of improving accuracy.
WiFi Sensing (Funded by TTDF Project from Department of Telecommunications)
Publications at IMWUT 2025
WiFi sensing has been studied over the past decade. However, such sensing consumes precious bandwidth, leading to weakening of data rates used for communication. To mitigate this problem, this project identifies the portions of the bandwidth used that contributes least to improvement in accuracy, and utilize these portions for data communication.
Indoor Visible Light Communication
Publications at Computer Communications 2024 and IEEE TVT 2025
With high data rates applications becoming common for indoor use, indoor visible light communication has been proposed to supplement WiFi communication. Such LiFi communication faces a number of challenges such as lower coverage, high handoffs and vulnerability to blockages. This project utilizes techniques of optimization to mitigate the above problems and make utilization of LiFi feasible for practical applications.