Recent Research Projects
Optimization and Pruning at Emerging Systems Lab, Intel Labs
Mentors : Nilesh Jain (Principal Engineer), Dr. Pablo Munoz, Vui Seng Chua
August 2022 - May 2023
Having worked majorly in domains of surveillance and computer vision previously, moving to optimization and pruning based projects was challenging but is proving a good steep learning curve. I realized that this knowledge and experience would help me a lot by imparting me with a lot of knowledge and corporate research experience in domain of systems which I could utilize to optimize my computer vision based models in the future. Currently I am working with Pablo on BoostrapNAS and heading my pruning based research project which is being advised by Nilesh, Vui Seng and Pablo.
Continual Learning at XuLab, Carnegie Mellon University
Mentors : Prof. Min Xu, Dr. Sima Behpour
June 2021 - February 2022
Continual Learning is perhaps one of the most important areas when it comes to deep learning. This is mainly because, achieving it with minimal catastrophic forgetting could help neural networks not just learn various classes more effectively but it could also put the AI community one step closer in the pursuit of Artificial General Intelligence. I had recently worked on a project based on it at the Xulab, CMU where upon collaboration with Prof. Min Xu and Dr. Sima Behpour, we had submitted our work recently to a top tier conference.
Pruning of Tracking and Detection models for Autonomous Vehicles at Optimization and Trustworthy ML Group, Michigan State University
Mentors : Prof. Sijia Liu
March 2022 - July 2022
Since the advent of deep learning, there has always been debates and discussions about the benefits of large networks, but also about the drawbacks it comes with, which are mostly space consumed and problems in deploying for real-world scenarios. Autonomous Vehicles is one such challenging real-world scenario as it requires most accurate models to operate in order to make it as safe as possible for both drivers as well as pedestrians. Currently, I am working with Prof. Sijia Liu on this where I am developing lightweight and accurate models for tracking and detection.
Helmet and Triple Riding Violations Detection at CVIT, International Institute of Information Technology, Hyderabad
Mentors : Prof. Ravi Kiran, Dr. Anbumani Subramanian, Prof. C.V. Jawahar
February 2021 - September 2021
Deep Learning and Computer Vision have always been strong contendors for getting utilized in road surveillance related applications. However, there exist problems such as joint detection and tracking of helmet and triple riding violations which have been mostly overlooked by researchers especially in crowded and unconstrained scenarios such as Indian roads. This problem statement becomes even more important given various reports from multiple sources such as WHO and World Bank which clearly indicate that a majority of road fatalities occur due to road crashes. Hence, I had proposed a novel solution which jointly detects as well as tracks helmet and triple riding violations. Our work is currently under review at a top tier conference. A patent is also being filed for the propose technique.