基于联邦强化学习的分布式模型剪枝  

Distributed Model Pruning Based on Federated Reinforcement Learning

作　　者：聂宇铭 臧文科 马学豪 刘宇儒 包致成 张镇 彭亿 NIE Yu-Ming;ZANG Wen-Ke;MA Xue-Hao;LIU Yu-Ru;BAO Zhi-Cheng;ZHANG Zhen;PENG Yi(China University of Petroleum,Qingdao 266580,China;Bureau of Industry and Information Technology of Qingdao West Coast New Area,Qingdao 266555,China)

机构地区：[1]中国石油大学(华东),青岛266580 [2]青岛西海岸新区工业和信息化局,青岛266555

出　　处：《计算机系统应用》2024年第5期154-161,共8页Computer Systems & Applications

摘　　要：联邦学习系统中,在资源受限的边缘端进行本地模型训练存在一定的挑战.计算、存储、能耗等方面的限制时刻影响着模型规模及效果.传统的联邦剪枝方法在联邦训练过程中对模型进行剪裁,但仍存在无法根据模型所处环境自适应修剪以及移除一些重要参数导致模型性能下降的情况.本文提出基于联邦强化学习的分布式模型剪枝方法以解决此问题.首先,将模型剪枝过程抽象化,建立马尔可夫决策过程,使用DQN算法构建通用强化剪枝模型,动态调整剪枝率,提高模型的泛化性能.其次设计针对稀疏模型的聚合方法,辅助强化泛化剪枝方法,更好地优化模型结构,降低模型的复杂度.最后,在多个公开数据集上将本方法与不同基线方法进行比较.实验结果表明,本文所提出的方法在保持模型效果的同时减少模型复杂度.There are challenges in training local models at resource-constrained edges in federated learning systems.The limitations in computing,storage,energy consumption,and other aspects constantly affect the scale and effectiveness of the model.Traditional federated pruning methods prune the model during the federated training process,but they fail to prune models adaptively according to the environment and may remove some important parameters,resulting in poor performance of models.This study proposes a distributed model pruning method based on federated reinforcement learning to solve this problem.Firstly,the model pruning process is abstracted,and a Markov decision process is established.DQN algorithm is used to construct a universal reinforcement pruning model,so as to dynamically adjust the pruning rate and improve model generalization performance.Secondly,an aggregation method for sparse models is designed to reinforce and generalize pruning methods,optimize the structure of the model,and reduce its complexity.Finally,this method is compared with different baselines on multiple publicly available datasets.The experimental results show that the proposed method maintains model effectiveness while reducing model complexity.

关 键 词：联邦学习 模型剪枝 强化学习 联邦剪枝 深度学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]