Computer Science Speaking Skills Talk

Monday, May 2, 2022 - 12:00pm to 1:00pm


In Person Traffic21 Classroom, Gates Hillman 6501


MICHAEL RUDOW, Ph.D. StudentComputer Science DepartmentCarnegie Mellon University

Learning-based Streaming Codes are Approximately Optimal for Variable-Size Frames

Providing a high quality-of-service for live communication is a pervasive challenge plagued by bursts of packet losses during transmission. Streaming codes are designed explicitly to recover lost packets in real-time for such low-latency streaming communication settings using minimal bandwidth. Applications such as live video streaming involve transmitting messages whose sizes vary over time due to compressing frames prior to transmission. The variability demands extra redundancy because consecutive large frames can be lost in a burst, leaving less bandwidth remaining for data. Mitigating the adverse effects of variability necessitates spreading the messages over multiple future packets. However, the optimal strategy for spreading depends on the sizes of future messages, which are not available due to streaming codes operating in an “online” setting. Algebraic coding techniques alone are hence insufficient for designing optimal streaming codes. We present a learning-based approach that combines machine learning with algebraic coding techniques to design the first approximately optimal streaming codes for a range of parameter regimes suitable for practical applications. Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.

Event Website:

For More Information, Contact:


Speaking Skills