高级搜索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

工业物联网中基于信息熵的联邦增量学习算法与优化

杨睿哲 谢欣儒 滕颖蕾 李萌 孙艳华 张大君

杨睿哲, 谢欣儒, 滕颖蕾, 李萌, 孙艳华, 张大君. 工业物联网中基于信息熵的联邦增量学习算法与优化[J]. 电子与信息学报. doi: 10.11999/JEIT231240
引用本文: 杨睿哲, 谢欣儒, 滕颖蕾, 李萌, 孙艳华, 张大君. 工业物联网中基于信息熵的联邦增量学习算法与优化[J]. 电子与信息学报. doi: 10.11999/JEIT231240
YANG Ruizhe, XIE Xinru, TENG Yinglei, LI Meng, SUN Yanhua, ZHANG Dajun. Entropy-based Federated Incremental Learning and Optimization in Industrial Internet of Things[J]. Journal of Electronics & Information Technology. doi: 10.11999/JEIT231240
Citation: YANG Ruizhe, XIE Xinru, TENG Yinglei, LI Meng, SUN Yanhua, ZHANG Dajun. Entropy-based Federated Incremental Learning and Optimization in Industrial Internet of Things[J]. Journal of Electronics & Information Technology. doi: 10.11999/JEIT231240

工业物联网中基于信息熵的联邦增量学习算法与优化

doi: 10.11999/JEIT231240
基金项目: 国家自然科学基金(62171062, 62371012)
详细信息
    作者简介:

    杨睿哲:女,副教授,研究方向为区块链技术、无线通信网络等

    谢欣儒:女,硕士生,研究方向为联邦学习、机器学习

    滕颖蕾:女,教授,研究方向为网络组网理论、随机优化等

    李萌:男,副教授,研究方向为边缘计算、无线通信网络等

    孙艳华:女,副教授,研究方向为机器学习、边缘计算等

    张大君:男,助理研究员,研究方向为人工智能,计算机、通信的智能网络等

    通讯作者:

    杨睿哲 yangruizhe@bjut.edu.cn

  • 中图分类号: TN919.8

Entropy-based Federated Incremental Learning and Optimization in Industrial Internet of Things

Funds: The National Natural Science Foundation of China (62171062, 62371012)
  • 摘要: 面对工业生产过程中大规模、多样且随时间增长的数据和机器学习任务,该文提出一种基于信息熵的联邦增量学习(FIL)与优化方法。基于联邦框架,各本地计算节点可利用本地数据进行模型训练,并计算信息平均熵上传至服务器,以此辅助识别类增任务;全局服务器则根据本地反馈的平均熵选择参与当前轮次训练的本地节点,并判决任务是否产生增量后,进行全局模型下发与聚合更新。所提方法结合平均熵和阈值进行不同情况下的节点选择,实现低平均熵下的模型稳定学习和高平均熵下的模型增量式扩展。在此基础上,采用凸优化,在资源有限的情况下自适应地调整聚合频率和资源分配,最终实现模型的有效收敛。仿真结果表明,在不同的情景下,该文所提方法都可以加速模型收敛并提升训练精度。
  • 图  1  工业场景中的系统模型图

    图  2  联邦式增量学习系统流程

    图  3  不同情况下4种算法的平均精度

    图  4  资源充足时,阈值对模型训练的影响

    图  5  资源充足时,数据量与节点类对模型训练的影响

    图  6  不同资源下,不同模型的训练情况

    表  1  符号描述

    符号 描述
    $ {\boldsymbol{\theta}} ^{r,t},{\boldsymbol{\theta}} _m^{\tau ,t} $ 全局模型及本地节点$m$处的局部模型
    $ f_{\rm c}^{}( \cdot ),f_{{\mathrm{d}}}^{}( \cdot ) $ 模型训练损失与蒸馏损失
    $ \mathcal{S}^r $ 第$ r $次迭代选择的节点集合
    $M,N$ 节点总数与每轮选中节点数
    $ \mathcal{H}_m^{r,t} $ 本地数据与模型间的信息熵
    $ \mathcal{I}( \cdot ) $ 熵函数
    $ {\varLambda _{{\mathrm{low}}}},{\varLambda _{{\mathrm{high}}}} $ 贪婪算法下节点选择的阈值
    $\eta $ 学习率
    $c^{\rm{loc}},c^{\rm{ent}}$ 本地节点训练任务负载
    $c^{\rm{glo}}$ 全局聚合复杂度
    $ t^*,E^* $ 模型迭代中的时延及能耗
    $R^{\rm{glo}},R^{\rm{loc}}$ 全局及本地训练迭代次数
    $f_m^{\rm{cmp}}$ 计算能力
    $ \varphi ,\beta ,\delta $ 凸优化过程中理论存在的中间值
    $ {\hat \varphi ^{r,t}} $,$ {\hat \beta ^{r,t}} $,${\hat \delta ^{r,t}}$ 凸优化过程中实测的中间值
    下载: 导出CSV

    1  基于信息熵的节点选择

     (1)输入:待选节点数$M$;输出:选中的节点集合${\mathcal{S}^r}$
     (2)初始化选择系数${\varLambda _{\rm{low}}}$和${\varLambda _{\rm{high}}}$,开始轮次$\mathcal{E}$
     (3)for $ r = 1,\; 2,\;\cdots ,\;{{\mathrm{epoch}}} $
     (4)  for $\forall m \in M$then利用式(6)计算$ \mathcal{H}_m^{r,t} $上传给CPU
     (5)  $ \mathcal{H}^{r,t} = \{ \mathcal{H}_m^{r,t}\} $,随机获取${r_{\rm{and}}} \in (0,1)$
     (6)  if $r < \mathcal{E}$ and $ {r_{\rm{and}}} < {\varLambda _{\rm{low}}} $
     (7)   then 随机选择$ N $个节点为$\mathcal{S}$, go to (10)
     (8)  if $ {\varLambda _{\rm{low}}} \le {r_{\rm{and}}} < {\varLambda _{\rm{high}}} $ then选$ \mathcal{H}_m^{r,t} $最小的$ N $个节点为${\mathcal{S}^r}$
     (9)   else:排序$ \mathcal{H}^{r,t} $,选$ \mathcal{H}_m^{r,t} $最大的$ N $个节点为${\mathcal{S}^r}$
     (10) 被选节点$ s_{m*}^t = 1 $, $ q_{m*}^r = 1 $
    下载: 导出CSV

    2  优化模型求解

     (1)初始化:$ {{\boldsymbol{\zeta}} ^{(0)}} $, $ v = 0 $
     (2)if $ {{\boldsymbol{\zeta}} ^{(v)}} $为稳态解, then返回$ {{\boldsymbol{\zeta}} ^{(v)}} $
     (3)$ \widehat{\boldsymbol{\zeta}}({\boldsymbol{\zeta}}^{(v)})\triangleq \mathrm{arg}\mathrm{min}\tilde{U}({\boldsymbol{\zeta}} ,{{\boldsymbol{\zeta}} }^{(v)}) $
     (4)$ {\boldsymbol{\zeta}}^{(v+1)}={\boldsymbol{\zeta}}^{(v)}+{k}^{(v)}{(}\widehat{\boldsymbol{\zeta}}({\boldsymbol{\zeta}}^{(v)})-{\boldsymbol{\zeta}}^{(v)}) $
     (5)$v \leftarrow v + 1$,转到(2)
    下载: 导出CSV

    表  2  参数说明

    参数 定义
    $ M $ 节点数量 [10, 200]
    $ E $ 总能量 [500, 1 000] J
    $t$ 总时间 [2 000, 20 000] s
    $ \left| {{\mathcal{D}_m}} \right| $ 节点$m$数据集大小 [120, 200]
    $ {c^{\rm{loc}}} $ 本地计算负荷 15 cycles
    $ {c^{\rm{glo}}} $ 全局计算负荷 25 cycles
    $ f_m^{\rm{cmp}} $ 节点$m$计算能力 [1, 10] GHz
    $ \psi /2 $ 节点计算芯片组的有效电容系数 $ {10^{ - 28}} $
    $p_m^{\min },p_m^{\max }$ 设备节点功率约束 [0.2, 1] W
    下载: 导出CSV
  • [1] BJORNSON E and SANGUINETTI L. Scalable cell-free massive MIMO systems[J]. IEEE Transactions on Communications, 2020, 68(7): 4247–4261. doi: 10.1109/tcomm.2020.2987311.
    [2] YU Wanke and ZHAO Chunhui. Broad convolutional neural network based industrial process fault diagnosis with incremental learning capability[J]. IEEE Transactions on Industrial Electronics, 2020, 67(6): 5081–5091. doi: 10.1109/tie.2019.2931255.
    [3] HUO Ru, ZENG Shiqin, WANG Zhihao, et al. A comprehensive survey on Blockchain in industrial internet of things: Motivations, research progresses, and future challenges[J]. IEEE Communications Surveys & Tutorials, 2022, 24(1): 88–122. doi: 10.1109/comst.2022.3141490.
    [4] SU Hang, QI Wen, HU Yingbai, et al. An incremental learning framework for human-like redundancy optimization of anthropomorphic manipulators[J]. IEEE Transactions on Industrial Informatics, 2022, 18(3): 1864–1872. doi: 10.1109/tii.2020.3036693.
    [5] KIRKPATRICK J, PASCANU R, RABINOWITZ N, et al. Overcoming catastrophic forgetting in neural networks[J]. Proceedings of the National Academy of Sciences of the United States of America, 2017, 114(13): 3521–3526. doi: 10.1073/pnas.1611835114.
    [6] SHOHAM N, AVIDOR T, KEREN A, et al. Overcoming forgetting in federated learning on non-IID data[J]. arXiv: 1910.07796, 2019. doi: 10.48550/arXiv.1910.07796.
    [7] LI Zhizhong and HOIEM D. Learning without forgetting[C]. The 14th European Conference on Computer Vision, Amsterdam, The Netherlands, 2016: 614–629. doi: 10.1007/978-3-319-46493-0_37.
    [8] DONG Jiahua, WANG Lixu, FANG Zhen, et al. Federated class-incremental learning[C]. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, USA, 2022: 10154–10163. doi: 10.1109/cvpr52688.2022.00992.
    [9] MAZUR M, PUSTELNIK Ł, KNOP S, et al. Target layer regularization for continual learning using Cramer-Wold distance[J]. Information Sciences, 2022, 609: 1369–1380. doi: 10.1016/j.ins.2022.07.085.
    [10] REBUFFI S-A, KOLESNIKOV A, SPERL G, et al. iCaRL: Incremental classifier and representation learning[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, USA, 2017: 5533–5542. doi: 10.1109/cvpr.2017.587.
    [11] WANG Qiang, LIU Jiayi, JI Zhong, et al. Hierarchical correlations replay for continual learning[J]. Knowledge-Based Systems, 2022, 250: 109052. doi: 10.1016/j.knosys.2022.109052.
    [12] JI Zhong, LI Jin, WANG Qiang, et al. Complementary calibration: Boosting general continual learning with collaborative distillation and self-supervision[J]. IEEE Transactions on Image Processing, 2023, 32: 657–667. doi: 10.1109/tip.2022.3230457.
    [13] HAO Meng, LI Hongwei, LUO Xizhao, et al. Efficient and privacy-enhanced federated learning for industrial artificial intelligence[J]. IEEE Transactions on Industrial Informatics, 2020, 16(10): 6532–6542. doi: 10.1109/tii.2019.2945367.
    [14] CHEN Zhixiong, YI Wenqiang, DENG Yansha, et al. Device scheduling for wireless federated learning with latency and representativity[C]. 2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME), Maldives, Maldives, 2022: 1–6. doi: 10.1109/iceccme55909.2022.9988590.
    [15] YANG Zhaohui, CHEN Mingzhe, SAAD W, et al. Energy efficient federated learning over wireless communication networks[J]. IEEE Transactions on Wireless Communications, 2021, 20(3): 1935–1949. doi: 10.1109/twc.2020.3037554.
    [16] DINH C T, TRAN N H, NGUYEN M N H, et al. Federated learning over wireless networks: Convergence analysis and resource allocation[J]. IEEE/ACM Transactions on Networking, 2021, 29(1): 398–409. doi: 10.1109/tnet.2020.3035770.
    [17] JING Shusen and XIAO Chengshan. Federated learning via over-the-air computation with statistical channel state information[J]. IEEE Transactions on Wireless Communications, 2022, 21(11): 9351–9365. doi: 10.1109/twc.2022.3175887.
    [18] YANG Ruizhe. The adaptive distributed learning based on homomorphic encryption and blockchain[J]. High Technology Letters, 2022, 28(4): 337–344. doi: 10.3772/j.issn.1006-6748.2022.04.001.
    [19] ZHAO Rui, SONG Jinming, YUAN Yufeng, et al. Maximum entropy population-based training for zero-shot human-AI coordination[C]. The AAAI Conference on Artificial Intelligence, Washington, USA, 2023: 6145–6153. doi: 10.1609/aaai.v37i5.25758.
    [20] ZHANG Peiying, WANG Chao, JIANG Chunxiao, et al. Deep reinforcement learning assisted federated learning algorithm for data management of IIoT[J]. IEEE Transactions on Industrial Informatics, 2021, 17(12): 8475–8484. doi: 10.1109/tii.2021.3064351.
    [21] WANG Shiqiang, TUOR T, SALONIDIS T, et al. Adaptive federated learning in resource constrained edge computing systems[J]. IEEE Journal on Selected Areas in Communications, 2019, 37(6): 1205–1221. doi: 10.1109/jsac.2019.2904348.
    [22] KRIZHEVSKY A and HINTON G. Learning multiple layers of features from tiny images[J]. Handbook of Systemic Autoimmune Diseases, 2009, 1(4).
    [23] SHAFIQ M and GU Zhaoquan. Deep residual learning for image recognition: A survey[J]. Applied Sciences, 2022, 12(18): 8972. doi: 10.3390/app12188972.
    [24] MINAEE S, KALCHBRENNER N, CAMBRIA E, et al. Deep learning-based text classification: A comprehensive review[J]. ACM Computing Surveys, 2022, 54(3): 62. doi: 10.1145/3439726.
  • 加载中
图(6) / 表(4)
计量
  • 文章访问数:  82
  • HTML全文浏览量:  42
  • PDF下载量:  13
  • 被引次数: 0
出版历程
  • 收稿日期:  2023-11-08
  • 修回日期:  2024-04-21
  • 网络出版日期:  2024-05-13

目录

    /

    返回文章
    返回