Conference Paper Metadata:
Title: "Download Speed Optimization in P2P Networks Using Decision Making and Adaptive Learning"
Authors: Aristeidis Karras, Christos Karras, Konstantinos C. Giotopoulos, Ioanna Giannoukou, Dimitrios Tsolis, Spyros Sioutas
Computer Engineering and Informatics Department, University of Patras
Paper URL: [ Ссылка ]
DOI: [ Ссылка ]
Abstract: Pure peer-to-peer networks serve to secure information in a decentralized, distributed topology. The multi-armed bandit (MAB) problem formulation proves to be a useful tool for analyzing the problem of optimizing new peer connections. In this paper, we outline the new peer scenario described as a reinforcement learning problem with MABs in order to identify the fastest peer to download from during the connection process. The MAB problem involves k slot machines which are also called one-armed bandits and pay out reward values according to an internal distribution, of which the agent is not aware. The aim is to choose a strategy to learn which arms pay out the most in order to maximize total reward over a set number of rounds. Results indicate that UCB and ε-first performed the best at selecting the optimal peer in each of our test scenarios. Contrariwise, SoftMax and ε-greedy unperformed.
Cite our work:
@InProceedings{10.1007/978-3-031-14054-9_22,
author="Karras, Aristeidis
and Karras, Christos
and Giotopoulos, Konstantinos C.
and Giannoukou, Ioanna
and Tsolis, Dimitrios
and Sioutas, Spyros",
editor="Daimi, Kevin
and Al Sadoon, Abeer",
title="Download Speed Optimization in P2P Networks Using Decision Making and Adaptive Learning",
booktitle="Proceedings of the ICR'22 International Conference on Innovations in Computing Research",
year="2022",
publisher="Springer International Publishing",
address="Cham",
pages="225--238",
isbn="978-3-031-14054-9"
}
The International Conference on Innovations in Computing Research
ICR 2022: Proceedings of the ICR’22 International Conference on Innovations in Computing Research pp 225–238
Keywords: #p2p #ai #machinelearning #network
