A Bayesian Approach to Decentralized Video Conferencing

Loading...
Thumbnail Image

Date

2023-05

Journal Title

Journal ISSN

Volume Title

Publisher

The Ohio State University

Research Projects

Organizational Units

Journal Issue

Abstract

Recent years have seen a significant increase in the adoption of video conferencing as a preferred means of communication. The centralized nature of the leading video conferencing platforms has brought with it various challenges and vulnerabilities. Previous research indicates that decentralized solutions offer improvements to centralized systems. This research continues past work by implementing DecVi++, a simulation for a novel protocol for providing decentralized video conferencing based on the Thompson sampling approach to the combinatorial multi-armed bandit problem, which can be used to model video conferencing clients and servers in a decentralized network. Thompson sampling, a Bayesian approach, allows for the addition of prior information into a system to improve future predictions for selecting a decentralized path for a video stream. An event-driven simulator was developed to test this protocol and compare it with the current leading video conferencing platform and a previous decentralized protocol, DecVi, which used a frequentist approach to the combinatorial multi-armed bandit problem. The simulator modeled the end devices of a video-conferencing network as nodes in a bipartite graph. Edges between nodes indicated the path of a video stream and weights on the edges were included as estimates of the bandwidth and latency of a given path. Using Thompson sampling, these values were sampled from normal distributions as inputs for the protocol to determine the optimal paths from a source client to recipient clients. Client feedback was then used to update the hyperparameters of these distributions. Experiments using the simulator showed that the incorporation of prior knowledge into the system produced more optimal video streaming paths. The DecVi++ algorithm tended to yield higher initial performance rewards than the DecVi algorithm but was sensitive to various hyperparameter initializations and experiment setups. The results of this research suggest that DecVi++ offers an encouraging approach to decentralized video conferencing.

Description

Keywords

computer networking

Citation