Plume:Lightweight and Generalized Congestion Control with Deep Reinforcement Learning

作     者：Dehui Wei Jiao Zhang Xuan Zhang Chengyuan Huang Dehui Wei;Jiao Zhang;Xuan Zhang;Chengyuan Huang

作者机构：State Key Laboratory of Networking and Switching TechnologyBUPT 100876China 

出 版 物：《China Communications》 (中国通信（英文版）)

年 卷 期：2022年第19卷第12期

页      面：101-117页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 081201[工学-计算机系统结构] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by National Natural Science Foundation of China (NSFC) under Grant (No.61872401) National Natural Science Foundation of China (NSFC) under Grant (No.62132022) Fok Ying Tung Education Foundation (No.171059) BUPT Excellent Ph.D.Students Foundation (No. CX2021102) 

主　　题：congestion control deep reinforcement learning lightweight generalization 

摘      要：Congestion control(CC)is always an important issue in the field of networking,and the enthusiasm for its research has never diminished in both academia and *** current years,due to the rapid development of machine learning(ML),the combination of reinforcement learning(RL)and CC has a striking ***,These complicated schemes lack generalization and are too heavyweight in storage and computing to be directly implemented in mobile *** order to address these problems,we propose Plume,a high-performance,lightweight and generalized RL-CC *** proposes a lightweight framework to reduce the overheads while preserving the original ***,Plume innovatively modifies the framework parameters of the reward function during the retraining process,so that the algorithm can be applied to a variety of *** results show that Plume can retain almost all the performance of the original model but the size and decision latency can be reduced by more than 50%and 20%,***,Plume has better performances in some special scenes.