CS Talk - Sohee Kim Park

Event time: 
Wednesday, January 12, 2022 - 4:00pm
Zoom Presentation See map
Event description: 

CS Talk
Sohee Kim Park

Host: Holly Rushmeier

Title: Advancing Quality of Experience of 360◦ Video Streaming using Machine Learning and Multipath
Network Protocol


Research interest has been significantly emerged in streaming 360◦ videos over the Internet to provide an immersive experience to the users. A major challenge of streaming such videos is that the amount of video data to be streamed could be an order of magnitude larger than conventional videos to achieve a similar viewing experience. The reason for this is that conventional streaming solutions fetch the entire panoramic scene, while the user views only a small part of the scene depending on his/her viewing direction in 360◦ videos.

In this talk, I will present my research on advancing the user quality of experience (QoE) on 360◦ video streaming using deep learning techniques and network protocol innovations. In the first part of the talk, I propose a system called Mosaic, which utilizes deep learning techniques to predict user’s viewing direction and uses a knapsack-based rate adaptation algorithm such that it maximizes the user-perceived video bitrates given available network capacity. In Mosaic+, I further improve the user QoE by utilizing a more representative metric of QoE as opposed to only the bitrates as in Mosaic.

In the second part of the talk, I will present a Reinforcement Learning (RL) based adaptive  streaming framework, called ATRIA, which further improves the adaptiveness to the network and user behaviors by exploiting the sequential decision-making nature of video streaming techniques. I will share how I address the challenges of using RL in this scenario, such as large action space and delayed reward evaluation.

In the last part of this talk, I will address the limited bandwidth issue by aggregating multiple network paths and using related transport layer innovations. With an early promise of my work that viewport prediction and rate adaptation work well for videos not used for training, I will conclude by describing avenues for future work, including improving the QoE of applications that offer immersive user experience with omnidirectional 360° images, unlocking the other bandwidth-hungry delay-sensitive applications to Machine Learning and Network Innovations.

Sohee Kim Park is a Vice President of Intelibs, Inc., an incubator company at CEWIT (Center of Excellence in Wireless and Information Technology) at Stony Brook University’s R&D Park. Before joining Intelibs, she worked for Verizon Wireless, Triveni Digital, Siemens Cooperate Research, and TongYang SHL System House. She received her Ph.D. in Computer Science at Stony Brook University. Previously, she received her M.S. and B.S. in Computer Science from Yale University and  Yonsei University, respectively.

Her research interests are designing and developing systems on theoretically sound principles, applying data science, machine learning, AI, and network protocol innovations to advance the user QoE and improve system performance over the wireless network. She emphasizes making such  solutions deployable and scalable with commercialization in mind. Her research is published in proceedings of the IEEE Sarnoff Symposium, IFIP Networking, IEEE WACV, and IEEE TNSM. As a seasoned entrepreneur, her extensive experiences in technology commercialization include distributed antenna systems deployed at multiple university campuses and hospitals and GPS products for nationwide wireless carriers.

She is passionate about preparing leaders for the next generation, developing technology that improves our quality of life, and partnering with Non-Profit Organizations to reach out to people in need.