Mitigation of Cold Start Problem in Experience-Based Adaptive Streaming over NDN

Abstract

Dynamic adaptive streaming over NDN typically relies on past information of network conditions and streaming quality. In this paper, we address the cold start problem associated with reinforcement learning based NDN adaptive streaming where a new consumer often found choosing bitrate arbitrarily due to the lack of experience. The idea is to construct a shared Q-Table which is continuously updated by previous consumers. Based on this Q-Table, a new consumer is expected to start choosing the segment bitrate more proactively. Simulations through ns-3 show that the proposed approach could help the consumers to find an optimal action from the beginning of the session.