Computer Science · Computer Networks and Communications
NAORL
Wei Zhang, Xuefeng Tao, Jianan Wang
See on the mapAbstract
Bandwidth Estimation(BWE) is the most important and challenging problem for Real Time Communication(RTC) systems. The rule-based BWE is designed with hand-crafted rules, which mainly depend on human knowledge, and therefore is difficult to generalize to unknown scenarios. Learning-based BWE algorithms, especially online reinforcement learning-based algorithms, are proposed to explore new decisions in complex network environments adaptively. However, these algorithms require frequent interactions with the environment, which would cause catastrophic experience for RTC users.