I am looking for self-motivated students to work with me at UChicago. Please apply here, and feel free to drop me an email with your CV.
The vision of my research is to optimize complex networked systems by machine learning techniques.
In my PhD thesis work, I have demonstrated that this approach can dramatically improve the quality of Internet applications. However, the potential of this confluence between systems and machine learning goes far beyond Internet applications.
Motivation: Today’s Internet is an "eyeball economy" driven by applications such as video streaming (YouTube) and Internet telephony (Skype). For the providers of these applications, the key to success is maintaining high user-perceived Quality of Experience (QoE).
Approach: My dissertation demonstrates that data-driven techniques can improve Internet QoE by utilizing a centralized real-time view of performance across millions of clients. My thesis work has three key pieces:
Conventional wisdom has been that TCP can ensure fairness among the video sessions sharing bottleneck links, but my research showed otherwise: these sessions could suffer from unfair, unstable quality, and inefficient bandwidth utilization, due to the behaviors of video streaming protocols. Building on this observation, I developed FESTIVE, the first HTTP-based video player that simultaneously improved fairness, stability, and efficiency over many commercial players. FESTIVE has inspired enormous work in industry and academia to further improve along these metrics.
Ecosystems of Internet applications are complex and federated. While any subsystem on the delivery chain, such as Comcast and Akamai, can impact end-to-end QoE, most of these subsystems cannot directly observe the impact on QoE as they change their configurations. This creates great room for improvement by aligning these subsystems with the end-to-end application QoE. Experience-Oriented Network Architecture (EONA) aims at bridging this gap by letting all subsystems along the delivery path be aware of the end-to-end QoE.