核化的多视角特权协同随机矢量功能链接网络及其增量学习方法

doi:10.13232/j.cnki.jnju.2022.02.011

南京大学学报(自然科学版) ›› 2022, Vol. 58 ›› Issue (2): 275–285.doi: 10.13232/j.cnki.jnju.2022.02.011

• • 上一篇

核化的多视角特权协同随机矢量功能链接网络及其增量学习方法

吴天宇, 王士同()

江南大学人工智能与计算机学院, 无锡, 214122

收稿日期:2021-09-24 出版日期:2022-03-30 发布日期:2022-04-02
通讯作者: 王士同 E-mail:wxjn00@163.com
作者简介:E⁃mail：wxjn00@163.com
基金资助:
国家自然科学基金(61972181)

Kernel Multi⁃view Privileged Random vector functional link net⁃work and its incremental learning method

Tianyu Wu, Shitong Wang()

School of Artificial Intelligence and Computer Science，Jiangnan University，Wuxi，214122，China

Received:2021-09-24 Online:2022-03-30 Published:2022-04-02
Contact: Shitong Wang E-mail:wxjn00@163.com

摘要/Abstract

摘要：

在许多实际应用场景中，可以从不同层次、不同角度获取相同对象的特征数据，如何有效地利用获取的多视角数据是一个值得研究的问题.和传统的单视角学习相比，多视角学习在多源数据的应用中显示了一定的优势.多角度学习（Multi?View Learning，MVL）面临的一个重要问题是在满足不同视角互补性的前提下如何保持视角之间的一致性.针对以上问题，提出一种新的多视角特权协同核化随机向量功能链接网络（KMPRVFL）来有效地解决多视角分类问题，其基本思想是将冗余视角的额外信息与平均视角上的特权信息相结合来监督当前视角的分类任务，将多视角数据用核化后加权线性组合成综合第二视角.同时，还设计了一种增量学习方法，可以有效地减少计算量.在真实数据集上的实验结果表明，和传统的多视角学习方法相比，KMPRVFL的能力更强，其平均测试精度要优于对比算法.

关键词: 多视角学习, 特权信息, 随机向量函数链接网络, 增量学习

Abstract:

In many practical application scenarios，how to effectively use the multi perspective data obtained from different levels and different angles to obtain the feature data of the same object is a problem worthy of study. Compared with traditional single perspective learning，multi perspective learning shows certain advantages in the ap?plication of multi?source data. An important problem in multi?view learning (MVL) is how to keep the consistency of perspectives while satisfying the complementarity of different perspectives. To solve problems above，a new kernel multi?view privileged random vector functional link network (KMPRVFL) is proposed to effectively solve the multi view classification problem. The basic idea is to combine the extra information of redundant perspective with the privileged information of average perspective to supervise the classification task of the current perspective. The multi?view data is combined into a comprehensive second view by weighted linear combination after kernel. At the same time，an incremental learning method is designed to effectively reduce the amount of calculation. Experimental results on real datasets show that KMPRVFL is more powerful than traditional multi?view learning methods. The average test accuracy of KMPRVFL algorithm is better than that of comparison methods.

Key words: MVL (Multi?View Learning), privileged information, RVFL (Random Vector Functional Link), incremental learning

中图分类号:

TP181

吴天宇, 王士同. 核化的多视角特权协同随机矢量功能链接网络及其增量学习方法[J]. 南京大学学报(自然科学版), 2022, 58(2): 275–285.

Tianyu Wu, Shitong Wang. Kernel Multi⁃view Privileged Random vector functional link net⁃work and its incremental learning method[J]. Journal of Nanjing University(Natural Sciences), 2022, 58(2): 275–285.

图/表 7

图1

图2

表1

图3

表2

表3

图4

参考文献 24

1	Yang Y， Wang H. Multi?view clustering：A survey. Big Data Mining and Analytics，2018，1(2)：83-107.
2	Chen Y F， Li D D， Zhang X，et al. Computer aided diagnosis of thyroid nodules based on the devised small?datasets multi?view ensemble learning. Medical Image Analysis，2021，67：101819.
3	洪思思，曹辰捷，王喆，等. 基于矩阵的AdaBoost多视角学习. 南京大学学报(自然科学)，2018，54(6)：1152-1160.
	Hong S S， Cao C J， Wang J. Matrix?based multi?view learning with AdaBoost. Journal of Nanjing University (Natural Science)，2018，54(6)：1152-1160.
4	Li J， Allinson N， Tao D C，et al. Multi?training support vector machine for image retrieval. IEEE Transactions on Image Processing，2006，15(11)：3597-3601.
5	Sun S L， Shawe?Taylor J. Sparse semi?supervised learning using conjugate functions. Journal of Machine Learning Research，2010(11)：2423-2455.
6	Sun S L. Multi?view Laplacian support vector machines∥Proceedings of the 7th International Conference on Advanced Data Mining and Applications. Springer Berlin Heidelberg，2011：209-222.
7	Vapnik V， Vashist A. A new learning paradigm：Learning using privileged information. Neural Networks，2009，22(5-6)：544-557.
8	Vapnik V， Izmailov R. Learning using privileged information：Similarity control and knowledge transfer. The Journal of Machine Learning Research，2015，16(1)：2023-2049.
9	Evers L， Messow C M. Sparse kernel methods for high?dimensional survival data. Bioinformatics，2008，24(14)：1632-1638.
10	Camoriano R. Large?scale kernel methods and applications to lifelong robot learning. 2019,arXiv:.
11	Sun S L， Jin F. Robust co?training. International Journal of Pattern Recognition and Artificial Intelligence，2011，25(7)：1113-1126.
12	Cheng J， Wang K Q. Multi?view sampling for relevance feedback in image retrieval∥18th International Conference on Pattern Recognition. Hong Kong，China：IEEE，2006，2：881-884.
13	张丹丹，邓赵红，蒋亦樟,等. 基于划分融合与视角加权的极大熵聚类算法. 计算机工程，2016，10(4)：554-564.
	Zhang D D， Deng Z H， Wang S T. Maximum entropy clustering algorithm for multi?view data. Journal of Frontiers of Computer Science and Technology，2016，10(4)：554-564.
14	Valente F， Wellekens C. Maximum entropy discrimination (MED) feature subset selection for speech recognition∥2003 IEEE Workshop on Automatic Speech Recognition and Understanding. St Thomas，VI，USA：IEEE，2003：327-332.
15	Sun S L， Chao G Q. Multi?view maximum entropy discrimination∥Proceedings of the 23rd International Joint Conference on Artificial Intelligence. Beijing，China：AAAI Press，2013：1706-1712.
16	Chao G Q， Sun S L. Consensus and complementarity based maximum entropy discrimination for multi?view classification. Information Sciences，2016(367-368)：296-310.
17	Tang J J， Tian Y J， Liu X H，et al. Improved multi?view privileged support vector machine. Neural Networks，2018(106)：96-109.
18	Tang J J， Tian Y J， Zhang P，et al. Multiview privileged support vector machines. IEEE Transactions on Neural Networks and Learning Systems，2018，29(8)：3463-3477.
19	Li Y M， Sun H J， Yan W Z，et al. R?CTSVM+：Robust capped L₁?norm twin support vector machine with privileged information. Information Sciences，2021(574)：12-32.
20	Pao Y H， Park G H， Sobajic D J. Learning and generalization characteristics of the random vector Functional?link net. Neurocomputing，1994，6(2)：163-180.
21	Bartlett P L. The sample complexity of pattern classification with neural networks ：the size of the weights is more important than the size of the network. IEEE Transactions on Information Theory，March 1998，44(2)：525-536.
22	Queipo N V， Haftka R T， Shyy W，et al. Surrogate?based analysis and optimization. Progress in Aerospace Sciences，2005，41(1)：1-28.
23	Chua T S， Tang J H， Hong R C，et al. Nus?wide：A real?world web image database from national university of Singapore∥Proceedings of the 8th ACM International Conference on Image and Video Retrieval. Santorini Island，Greece：ACM，2009：1-9.
24	Xian Y Q， Lampert C H， Schiele B，et al. Zero?shot learning:A comprehensive evaluation of the good，the bad and the ugly. IEEE Transactions on Pattern Analysis and Machine Intelligence，2019，41(9)：2251-2265.

相关文章 15

[1]	杨梅, 曾雯喜, 方宇, 闵帆. 多示例学习的两阶段实例选择和自适应包映射算法[J]. 南京大学学报(自然科学版), 2022, 58(1): 94-102.
[2]	卢舜, 林耀进, 吴镒潾, 包丰浩, 王晨曦. 基于多粒度一致性邻域的多标记特征选择[J]. 南京大学学报(自然科学版), 2022, 58(1): 60-70.
[3]	吕亚兰, 徐媛媛, 张恒汝. 一种可解释性泛化矩阵分解推荐算法[J]. 南京大学学报(自然科学版), 2022, 58(1): 135-142.
[4]	高菲, 杨柳, 李晖. 开放集识别研究综述[J]. 南京大学学报(自然科学版), 2022, 58(1): 115-134.
[5]	许国强, 余长州, 王林, 周春蕾, 高阳. 一种增强贝叶斯网络结构学习的自动变量序调整算法[J]. 南京大学学报(自然科学版), 2021, 57(2): 255-261.
[6]	周小亮, 吴东洋, 曹磊, 王玉鹏, 业宁. 基于修剪树的优化聚类中心算法[J]. 南京大学学报(自然科学版), 2021, 57(2): 167-176.
[7]	汪敏,赵飞,闵帆. 储层预测的代价敏感主动学习算法[J]. 南京大学学报(自然科学版), 2020, 56(4): 561-569.
[8]	朱荀,刘国强,丁华平,沈庆宏. 一种通过支持向量机对交通拥堵情况进行分类的方法[J]. 南京大学学报(自然科学版), 2020, 56(2): 278-283.
[9]	王卫星,刘兆伟,石敬华. 基于时间敏感滑动窗口的CP⁃nets结构学习[J]. 南京大学学报(自然科学版), 2020, 56(2): 175-185.
[10]	信统昌,刘兆伟. 基于贝叶斯⁃遗传算法的多值无环CP⁃nets学习[J]. 南京大学学报(自然科学版), 2020, 56(1): 74-84.
[11]	郑文萍,刘韶倩,穆俊芳. 一种基于相对熵的随机游走相似性度量模型[J]. 南京大学学报(自然科学版), 2019, 55(6): 984-999.
[12]	黄华娟,韦修喜. 基于自适应调节极大熵的孪生支持向量回归机[J]. 南京大学学报(自然科学版), 2019, 55(6): 1030-1039.
[13]	刘　素, 刘惊雷. 基于特征选择的CP－nets结构学习[J]. 南京大学学报(自然科学版), 2019, 55(1): 14-28.
[14]	贾海宁, 王士同. 面向重尾噪声的模糊规则模型[J]. 南京大学学报(自然科学版), 2019, 55(1): 61-72.
[15]	曲昭伟, 杨凯翔, 王晓茹, 李百威. 基于语义行为和社交关联的好友推荐模型[J]. 南京大学学报(自然科学版), 2018, 54(6): 1124-1131.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

数据集	数量	类别数	特征1	特征2	特征3	特征4
NUS?wide	4089	7	CM55 (225)	WT (128)	EDH (73)	COPR (144)
AwA2	5560	6	SURF (2000)	HOG(252)	CHF (2600)	ISS (2000)

	Datasets?A	Datasets?B	KMPRVFL		KRVFL		MED?2C		PSVM?2V
	Datasets?A	Datasets?B	Accuracy	STD	Accuracy	STD	Accuracy	STD	Accuracy	STD
	Average		84.00%	0.023	77.63%	0.018	78.23%	0.221	79.66%	0.024
1	buildings	computer	83.86%	0.021	71.93%	0.015	77.17%	0.015	78.36%	0.012
2	buildings	elk	86.57%	0.022	81.01%	0.017	81.04%	0.017	82.23%	0.015
3	buildings	fox	90.64%	0.013	82.28%	0.035	83.41%	0.035	84.30%	0.029
4	buildings	horses	85.35%	0.030	76.42%	0.011	78.87%	0.011	79.82%	0.015
5	buildings	moon	83.59%	0.023	75.16%	0.009	81.65%	0.009	78.52%	0.020
6	buildings	plants	85.98%	0.015	79.19%	0.013	81.34%	0.013	82.25%	0.007
7	buildings	road	72.68%	0.013	63.93%	0.026	62.08%	0.026	65.22%	0.027
8	computer	elk	86.12%	0.033	77.21%	0.025	77.47%	0.025	78.66%	0.025
9	computer	fox	85.41%	0.031	76.39%	0.030	75.18%	0.030	76.92%	0.024
10	computer	horses	89.04%	0.022	78.11%	0.012	81.52%	0.012	82.90%	0.016
11	computer	moon	83.10%	0.028	76.45%	0.026	74.64%	0.026	77.99%	0.029
12	computer	plants	87.00%	0.012	80.68%	0.014	79.90%	0.014	81.34%	0.013
13	computer	road	81.36%	0.021	74.51%	0.034	70.62%	0.034	76.17%	0.033
14	elk	fox	75.45%	0.030	68.95%	0.023	68.57%	0.023	70.65%	0.026
15	elk	horses	78.88%	0.021	76.67%	0.023	75.73%	0.023	76.74%	0.024
16	elk	moon	87.77%	0.029	82.68%	0.013	83.98%	0.013	84.53%	0.025
17	elk	plants	84.77%	0.033	82.21%	0.009	83.75%	0.009	83.09%	0.012
18	elk	road	82.60%	0.022	78.90%	0.032	79.24%	0.032	79.28%	0.032
19	fox	horses	83.80%	0.021	76.65%	0.018	77.68%	0.018	79.99%	0.019
20	fox	moon	86.22%	0.023	81.94%	0.044	82.25%	0.044	83.92%	0.045
21	fox	plants	79.83%	0.057	79.73%	0.014	73.88%	0.014	82.31%	0.004
22	fox	road	84.16%	0.033	78.81%	0.033	81.67%	0.033	79.65%	0.031
23	horses	moon	88.18%	0.033	82.56%	0.028	83.76%	0.028	84.59%	0.032
24	horses	plants	89.15%	0.011	85.05%	0.012	85.12%	0.012	86.73%	0.015
25	horses	road	80.14%	0.012	73.94%	0.035	74.57%	0.035	75.67%	0.039
26	moon	plants	86.90%	0.028	81.11%	0.033	82.40%	0.033	83.03%	0.027
27	moon	road	81.34%	0.027	74.71%	0.015	75.23%	0.015	76.67%	0.029
28	plants	road	82.13%	0.017	76.43%	0.031	77.68%	0.030	78.97%	0.030

	Datasets?A	Datasets?B	KMPRVFL		KRVFL		MED?2C		PSVM?2V
	Datasets?A	Datasets?B	Accuracy	STD	Accuracy	STD	Accuracy	STD	Accuracy	STD
	Average		87.41%	0.019	82.58%	0.031	79.78%	0.034	83.59%	0.054
1	chimps	panda	91.01%	0.009	87.14%	0.027	86.93%	0.027	90.92%	0.023
2	chimps	leopard	92.50%	0.01	86.84%	0.018	82.80%	0.04	87.61%	0.043
3	chimps	cat	92.18%	0.023	86.65%	0.034	82.07%	0.049	86.52%	0.067
4	chimps	pig	89.15%	0.05	82.09%	0.05	82.86%	0.033	83.81%	0.056
5	chimps	hippo	89.05%	0.014	86.84%	0.022	82.82%	0.043	84.45%	0.063
6	chimps	rat	89.78%	0.023	82.36%	0.041	75.12%	0.035	81.16%	0.082
7	chimps	seal	91.90%	0.012	81.46%	0.035	83.07%	0.026	87.76%	0.016
8	panda	leopard	92.83%	0.016	88.84%	0.014	84.31%	0.026	90.46%	0.012
9	panda	cat	95.05%	0.015	90.04%	0.026	88.03%	0.02	89.53%	0.067
10	panda	pig	90.53%	0.014	83.62%	0.025	78.83%	0.042	83.46%	0.035
11	panda	hippo	93.89%	0.013	87.69%	0.018	87.47%	0.009	91.13%	0.05
12	panda	rat	91.75%	0.02	85.92%	0.043	82.25%	0.025	86.57%	0.064
13	panda	seal	92.62%	0.024	89.10%	0.017	86.89%	0.033	89.63%	0.026
14	leopard	cat	91.74%	0.015	86.94%	0.021	86.15%	0.033	90.69%	0.045
15	leopard	pig	85.56%	0.037	82.62%	0.028	78.53%	0.027	84.11%	0.057
16	leopard	hippo	91.13%	0.01	83.10%	0.028	82.54%	0.037	88.49%	0.092
17	leopard	rat	87.58%	0.014	84.53%	0.033	80.17%	0.04	85.21%	0.082
18	leopard	seal	90.78%	0.019	86.47%	0.027	87.83%	0.045	88.00%	0.091
19	cat	pig	84.87%	0.037	79.27%	0.044	73.92%	0.061	76.80%	0.055
20	cat	hippo	89.57%	0.016	87.08%	0.029	85.07%	0.037	86.29%	0.074
21	cat	rat	77.01%	0.029	74.17%	0.04	62.40%	0.025	68.46%	0.043
22	cat	seal	86.52%	0.023	76.55%	0.044	82.60%	0.038	83.68%	0.06
23	pig	hippo	82.20%	0.019	77.21%	0.055	71.42%	0.037	74.49%	0.066
24	pig	rat	73.18%	0.029	69.21%	0.032	70.52%	0.025	74.31%	0.08
25	pig	seal	84.59%	0.021	78.27%	0.045	71.86%	0.037	77.08%	0.092
26	hippo	rat	85.17%	0.02	82.92%	0.019	72.94%	0.035	75.23%	0.056
27	hippo	seal	81.41%	0.026	70.45%	0.034	67.08%	0.03	69.48%	0.053
28	rat	seal	80.59%	0.036	77.32%	0.042	71.96%	0.02	75.69%	0.017

核化的多视角特权协同随机矢量功能链接网络及其增量学习方法

Kernel Multi⁃view Privileged Random vector functional link net⁃work and its incremental learning method

RichHTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 24

相关文章 15

Metrics

本文评价

推荐阅读 0