多示例嵌入学习的实例关联性挖掘与强化

doi:10.6040/j.issn.1671-9352.4.2022.606

Abstract

Abstract:

We propose the multi-instance embedding learning with instance affinity mining and reinforcement(MEMR) algorithm, including three techniques. The affinity mining technique is based on a custom affinity metric. First, the initial negative representative instance set(INRI) is selected in the negative instance space. Then, the initial positive representative instance set(IPRI) is chosen according to the difference between positive and negative instances. The affinity reinforcement technique evaluates the positive(negative) affinity between IPRI(INRI) and the entire instance space to obtain a representative instance set with stronger overall affinity. The bag embedding technique converts bags into single vectors for learning through the designed embedding function. Experiments are carried out across four application domains and seven comparison algorithms. The results show that MEMR generally outperforms other comparison algorithms in accuracy, especially in image retrieval and web recommendation datasets.

Key words: affinity mining, affinity reinforcement, embedding method, instance selection, multi-instance learning

CLC Number:

O213

Mei YANG,Wen DENG,Benwen ZHANG,Fan MIN. Multi-instance embedding learning with instance affinity mining and reinforcement[J].JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE), 2024, 59(1): 35-45.

Figures/Tables 12

Fig.1

Table 1

Fig.2

Table 2

Fig.3

Fig.4

Table 3

Table 4

Average accuracy on text categorization datasets"

Dataset	Simple-MI	MILFM	miFV	miVLAD	MILDM	Stable-MI	ELDB	MEMR
News.aa	83.6±0.80	52.6±7.86	83.8±1.60	84.0±2.28	54.6±7.06	52.6±4.03	84.6±1.78	86.0±2.28
News.cg	78.0±0.63	54.6±1.20	80.2±0.98	79.6±0.80	53.2±5.31	50.2±5.11	79.5±2.42	80.8±1.33
News.co	57.4±3.56	49.6±2.24	72.2±1.33	69.2±1.60	52.2±4.31	47.4±4.03	63.1±3.07	71.6±2.06
News.csi	75.4±0.80	57.6±3.20	79.8±1.17	80.0±1.55	56.6±6.62	50.2±5.49	78.1±2.02	81.2±2.40
News.csm	77.8±0.75	52.8±6.79	77.2±0.75	78.0±1.10	43.4±3.38	51.0±5.02	76.4±3.89	80.6±0.80
News.cw	71.0±3.16	57.8±2.79	86.6±0.80	82.6±1.02	56.8±4.31	54.2±4.49	79.6±1.43	81.0±1.10
News.mf	58.8±0.98	51.2±2.32	71.0±1.26	72.2±1.94	46.8±2.71	52.6±6.65	64.4±2.37	74.0±2.68
News.ra	75.4±0.49	52.4±1.62	78.4±1.36	81.6±1.02	51.8±6.68	52.0±5.02	71.5±2.27	82.4±1.74
News.rm	77.2±2.56	54.8±3.25	85.6±2.58	82.8±0.75	57.0±5.10	54.0±1.90	81.7±1.83	83.2±1.17
News.rsb	74.6±1.02	54.6±3.44	84.8±0.40	83.2±0.75	48.2±3.43	54.2±3.06	79.2±3.33	83.6±1.36
News.rsh	80.8±0.98	50.4±0.49	87.8±1.33	89.6±1.02	47.4±5.85	51.0±4.77	77.2±3.19	90.0±1.55
News.sc	73.8±0.40	58.6±1.85	75.2±1.60	83.0±1.10	49.4±3.83	50.2±4.53	70.0±2.87	82.8±1.60
News.se	92.0±0.00	53.0±0.00	92.6±0.80	92.4±0.49	55.6±1.96	51.0±3.58	88.2±1.32	92.4±1.36
News.sm	72.4±1.36	57.2±0.75	83.2±1.72	81.0±1.10	52.6±4.96	51.0±7.16	80.9±2.81	83.6±1.36
News.sr	77.4±0.80	50.2±1.17	79.8±2.56	79.8±2.32	51.4±2.80	57.6±2.65	80.7±1.16	80.6±1.62
News.ss	82.2±0.40	54.2±1.72	87.2±1.17	85.6±1.20	50.2±2.79	50.0±1.41	78.8±1.75	88.6±1.20
News.tpg	77.2±1.17	52.6±1.36	77.8±1.17	81.8±0.75	44.0±4.77	51.0±2.61	75.4±2.91	81.0±1.67
News.tpmd	83.0±1.79	60.0±3.58	79.0±0.63	83.6±1.02	55.6±2.94	55.6±4.59	76.7±2.26	85.2±1.17
News.tpmc	66.2±3.82	62.4±1.02	75.8±1.47	76.0±1.67	53.6±4.03	56.8±2.99	65.5±1.78	77.2±0.75
News.trm	61.6±1.02	52.6±1.02	75.0±1.10	78.0±2.28	47.2±3.87	51.0±3.63	66.4±2.27	76.2±1.47

Table 4

Table 5

Table 6

Fig.5

Table 7

References 33

1	DIETTERICH T G , LATHROP R H , LOZANO-PEREZ T . Solving the multiple instance problem with axis-parallel rectangles[J]. Artificial Intelligence, 1997, 89 (1/2): 31- 71.
2	LI Daxiang , ZHANG Yue . Multi-instance learning algorithm based on LSTM for Chinese painting image classification[J]. IEEE Access, 2020, 8, 179336- 179345. doi: 10.1109/ACCESS.2020.3027982
3	SHARMA Y, SHRIVASTAVA A, EHSAN L, et al. Cluster-to-conquer: a framework for end-to-end multi-instance learning for whole slide image classification[C]//Medical Imaging with Deep Learning. Lübeck: PMLR, 2021: 682-698.
4	王刚, 许信顺. 一种新的基于多示例学习的场景分类方法[J]. 山东大学学报(理学版), 2010, 45 (7): 108- 113.
	WANG Gang , XU Xinshun . A new multi-instance learning method for scene classification[J]. Journal of Shandong University (Natural Science), 2010, 45 (7): 108- 113.
5	YI Lin , ZHANG Honggang . Regularized instance embedding for deep multi-instance learning[J]. Applied Sciences, 2020, 10 (1): 64- 78.
6	KUMAR V, CHEMMENGATH S, GUPTA Y, et al. Multi-instance training for question answering across table and linked text[J/OL]. arXiv, 2021. https://arxiv.org/abs/2112.07337.
7	YANG Mei, ZENG Wenxi, MIN Fan. Multi-instance embedding learning through high-level instance selection[C]//Pacific-Asia Conference on Knowledge Discovery and Data Mining. Chengdu: Springer, 2022: 122-133.
8	TIAN Yuchi , WANG Jiawei , YANG Wenjie , et al. Deep multi-instance transfer learning for pneumothorax classification in Chest X-Ray images[J]. Medical Physics, 2022, 49 (1): 231- 243. doi: 10.1002/mp.15328
9	MANIVANNAN S , COBB C , BURGESS S , et al. Subcategory classifiers for multiple-instance learning and its application to retinal nerve fiber layer visibility classification[J]. IEEE Transactions on Medical Imaging, 2017, 36 (5): 1140- 1150. doi: 10.1109/TMI.2017.2653623
10	WEI Xiushen , YE Hanjia , MU Xin , et al. Multi-instance learning with emerging novel class[J]. IEEE Transactions on Knowledge and Data Engineering, 2019, 33 (5): 2109- 2120.
11	ZHANG Yalin, ZHOU Zhihua. Multi-instance learning with key instance shift[C]//International Joint Conference on Artificial Intelligence. Melbourne: IJCAI, 2017: 3441-3447.
12	XU Xin, FRANK E. Logistic regression and boosting for labeled bags of instances[C]//Pacific-Asia Conference on Knowledge Discovery and Data Mining. Berlin: Springer, 2004: 272-281.
13	MELKI G , CANO A , VENTURA S . MIRSVM: multi-instance support vector machine with bag representatives[J]. Pattern Recognition, 2018, 79, 228- 241. doi: 10.1016/j.patcog.2018.02.007
14	LUAN Tianxiang , LUO Tingjin , ZHUGE Wenzhang , et al. Optimal representative distribution margin machine for multi-instance learning[J]. IEEE Access, 2020, 8, 74864- 74874. doi: 10.1109/ACCESS.2020.2988764
15	CHIKONTWE P, KIM M, NAM S J, et al. Multiple instance learning with center embeddings for histopathology classification[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Lima: Springer, 2020: 519-528.
16	YANG Mei , ZHANG Yuxuan , WANG Xizhao , et al. Multi-instance ensemble learning with discriminative bags[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2021, 52 (9): 5456- 5467.
17	AMORES J . Multiple instance classification: review, taxonomy and comparative study[J]. Artificial Intelligence, 2013, 201, 81- 105. doi: 10.1016/j.artint.2013.06.003
18	WEI Xiushen, WU Jianxin, ZHOU Zhihua. Scalable multi-instance learning[C]//International Conference on Data Mining. Shenzhen: IEEE, 2014: 1037-1042.
19	ZHANG Minling , ZHOU Zhihua . Multi-instance clustering with applications to multi-instance prediction[J]. Applied Intelligence, 2009, 31 (1): 47- 68. doi: 10.1007/s10489-007-0111-x
20	CHEN Yixin , BI Jinbo , WANG J Z . MILES: multiple-instance learning via embedded instance selection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28 (12): 1931- 1947. doi: 10.1109/TPAMI.2006.248
21	LI Wujun . MILD: multiple-instance learning via disambiguation[J]. IEEE Transactions on Knowledge and Data Engineering, 2009, 22 (1): 76- 89.
22	HONG Richang , WANG Meng , GAO Yue , et al. Image annotation by multiple-instance learning with discriminative feature mapping and selection[J]. IEEE Transactions on Cybernetics, 2013, 44 (5): 669- 680.
23	WEI Xiushen , WU Jianxin , ZHOU Zhihua . Scalable algorithms for multi-instance learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2017, 28 (4): 975- 987. doi: 10.1109/TNNLS.2016.2519102
24	WU Jia , PAN Shirui , ZHU Xingquan , et al. Multi-instance learning with discriminative bag mapping[J]. IEEE Transactions on Knowledge and Data Engineering, 2018, 30 (6): 1065- 1080. doi: 10.1109/TKDE.2017.2788430
25	ZHANG Weijia, LI Jiuyong, LIU Lin. Robust multi-instance learning with stable instances[J/OL]. arXiv, 2019. https://arxiv.org/pdf/1902.05066.pdf.
26	FU Zhouyu , ROBLES-KELLY A , ZHOU Jun . MILIS: multiple instance learning with instance selection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33 (5): 958- 977.
27	MIN Fan , ZHANG Shiming , CIUCCI D , et al. Three-way active learning through clustering selection[J]. International Journal of Machine Learning and Cybernetics, 2020, 11 (5): 1033- 1046. doi: 10.1007/s13042-020-01099-2
28	ANDREWS S , TSOCHANTARIDIS I , HOFMANN T . Support vector machines for multiple-instance learning[J]. Neural Information Processing Systems, 2002, 14, 561- 568.
29	DECENCIERE E , ZHANG Xiwei , CAZUGUEL G , et al. Feedback on a publicly distributed image database: the Messidor database[J]. Image Analysis and Stereology, 2014, 33 (3): 231- 234. doi: 10.5566/ias.1155
30	KANDEMIR M , HAMPRECHT F A . Computer-aided diagnosis from weak supervision: a benchmarking study[J]. Computerized Medical Imaging and Graphics, 2015, 42, 44- 50. doi: 10.1016/j.compmedimag.2014.11.010
31	ZHOU Zhihua, SUN Yuyin, LI Yufeng. Multi-instance learning by treating instances as non-iid samples[C]//Proceedings of the 26th Annual International Conference on Machine Learning. Montreal: ACM, 2009: 1249-1256.
32	XU Bicun, TING Kaiming, ZHOU Zhihua. Isolation set-kernel and its application to multi-instance learning[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Anchorage: ACM, 2019: 941-949.
33	DEMSAR J . Statistical comparisons of classifiers over multiple data sets[J]. The Journal of Machine Learning Research, 2006, 7, 1- 30.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

Comments

Recommended 10

[1]	ZOU Guo-ping1, MA Ru-ning1, DING Jun-di2, ZHONG Bao-jiang3. Image retrieval based on saliency weighted color and texture[J]. J4, 2010, 45(7): 81 -85 .
[2]	CHEN Li . The robust fault diagnosis filter design for uncertain singular systems[J]. J4, 2007, 42(7): 62 -65 .
[3]	GUO Lan-lan1,2, GENG Jie1, SHI Shuo1,3, YUAN Fei1, LEI Li1, DU Guang-sheng1*. Computing research of the water hammer pressure in the process of #br# the variable speed closure of valve based on UDF method[J]. JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE), 2014, 49(03): 27 -30 .
[4]	SHI Kai-quan. P-information law intelligent fusion and soft information #br# image intelligent generation[J]. JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE), 2014, 49(04): 1 -17 .
[5]	ZHANG Ling ,ZHOU De-qun . Research on the relationships among the λ fuzzy measures, Mbius representation and interaction representation[J]. J4, 2007, 42(7): 33 -37 .
[6]	DING Chao1, 2, YUAN Chang-an1, 3, QIN Xiao1, 3. A prediction algorithm for multi-data streams based on GEP[J]. J4, 2010, 45(7): 50 -54 .
[7]	ZHANG De-yu,ZHAI Wen-guang . [J]. J4, 2006, 41(5): 4 -07 .
[8]	YANG Jianhui1, ZHANG Jianping, CHENG Xinlu, YANG Xiangdong. Magnetic dipole transitions among 1s²2s2p ³P_0,of Be-like ions[J]. J4, 2009, 44(11): 29 -34 .
[9]	PANG Yong,MU Zong-zhao,WANG Yue-hai,BAO Yu-hai,SUN Lei . Transpiration characteristics and its influencing factors of sixteen polar clones[J]. J4, 2006, 41(6): 168 -172 .
[10]	MENG Xiang-bo1, ZHANG Li-dong1, DU Zi-ping2. Investment and reinsurance strategy for insurers under #br# mean-variance criterion with jumps#br#[J]. JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE), 2014, 49(05): 36 -40 .

符号	含义	符号	含义
$\mathscr{X}$	实例空间	V_i	B_i的嵌入向量
$\mathscr{T}$	数据集	N	$\mathscr{T}$中包的个数
Y	标签向量	n_i	B_i中实例的个数
B_i	第i个包	d	实例的维度
x_ij	B_i的第j个实例	m	正包的个数
y_i	B_i的标签	ψ	C中代表实例的个数
C	代表实例集

数据集	包数量			实例数	维度
数据集	正包	负包	包	实例数	维度
Elephant	100	100	200	1 391	230
Fox	100	100	200	1 320	230
Tiger	100	100	200	1 220	230
Messidor	654	546	1 200	12 352	687
Ucsb_breast	26	32	58	2 002	708
Newsgroups	993	1 007	2 000	80 137	200
Web	490	527	1 017	30 807	6 211

Dataset	Simple-MI	MILFM	miFV	miVLAD	MILDM	Stable-MIL	ELDB	MEMR
Elephant	82.5±0.84	81.5±1.22	86.0±1.76	84.7±0.98	76.5±1.64	63.2±2.58	75.4±1.99	87.1±1.28
Fox	61.9±0.86	60.8±2.60	61.2±0.75	63.3±1.75	54.2±3.47	59.7±4.33	58.8±2.66	64.6±0.86
Tiger	81.1±1.16	76.3±1.29	79.1±0.58	84.9±7.63	69.0±1.41	65.7±2.06	67.4±2.73	85.1±0.37
Messidor	61.8±0.84	62.1±0.53	70.5±0.53	67.5±0.28	64.0±0.24	62.2±0.47	56.8±1.53	69.3±0.30
Ucsb_breast	81.2±2.71	55.6±2.33	85.6±0.80	80.0±1.79	56.0±2.19	54.4±0.20	63.0±7.62	82.8±4.12

Dataset	Simple-MI	MILFM	miFV	miVLAD	MILDM	Stable-MI	ELDB	MEMR
Web1	80.7±2.53	81.6±0.86	83.4±1.06	79.6±1.09	83.6±1.15	83.0±1.23	81.1±1.72	84.0±1.36
Web2	82.3±2.34	81.0±0.36	83.0±0.93	80.0±1.00	82.7±0.81	82.7±1.00	74.2±2.31	81.4±1.56
Web3	80.7±2.90	81.9±1.68	82.1±0.93	81.2±1.87	81.6±0.73	81.4±1.23	79.0±1.99	82.0±1.45
Web4	80.9±1.00	79.8±2.26	80.3±1.09	83.8±0.36	78.7±1.36	77.6±0.45	79.1±2.59	85.6±1.56
Web5	78.1±1.52	77.8±2.41	78.1±1.15	82.5±0.89	79.0±1.41	78.1±0.61	74.1±2.82	83.2±1.09
Web6	79.8±2.90	82.0±1.34	77.8±0.45	84.9±0.93	82.7±1.52	76.5±0.68	79.7±1.77	86.7±1.87
Web7	64.1±0.73	60.0±1.99	68.0±2.02	72.9±2.18	61.8±4.53	61.0±3.06	52.6±3.49	74.5±1.41
Web8	64.7±1.45	61.6±2.66	71.2±1.59	76.3±2.70	56.1±2.02	59.0±3.19	48.0±3.16	78.9±2.10
Web9	68.0±2.53	57.4±3.01	74.0±4.05	77.2±1.15	56.7±2.34	54.9±3.73	46.5±2.93	81.2±2.97

Datasets	Simple-MI	MILFM	miFV	miVLAD	MILDM	Stable-MIL	ELDB	MEMR
图像检索	3.33	5.00	3.33	2.33	6.67	7.33	7.00	1.00
医学图像	5.00	6.50	1.00	3.50	5.00	6.50	6.50	2.00
文本分类	4.45	6.45	2.55	2.25	7.25	7.30	4.20	1.55
网页推荐	4.89	5.11	3.33	3.67	4.33	5.78	7.33	1.56
平均排名	4.50	5.97	2.74	2.71	6.29	6.85	5.41	1.53

Multi-instance embedding learning with instance affinity mining and reinforcement

RichHTML

PDF (PC)

Abstract

Cite this article

share this article

Figures/Tables 12

References 33

Related Articles 2

Metrics

Comments

Recommended 10

Datasets(d/n/N)			Simple-MI		MILFM		miFV		miVLAD		MILDM		Stable-MIL		ELDB		MEMR
Time Complexity			O(dN)		O(dn²)		O(dn)		O(dn)		O(dn²)		O(dn²)		O(dn²)		O(dn)
Elephant	(230/1 320/200)		0.141		39.974		5.561		1.062		34.179		16.962		24.854		5.234
Ucsb_breast	(708/2 002/58)		0.125		83.145		8.576		1.390		65.969		152.230		63.860		4.624
News.aa	(200/5 443/100)		0.078		446.780		5.717		1.250		382.254		66.806		349.120		3.905
News.sm	(200/3 094/100)		0.063		130.253		5.030		1.094		121.533		40.662		114.382		3.406
Web4	(6 059/3 423/113)		1.109		533.881		711.693		9.748		453.940		1 169.954		785.484		44.241
Mean rank		1.00		7.40		4.00		2.00		6.40		6.20		6.00		3.00

[1]	Mei YANG,Wenjing KE,Dandong WANG. Feasible region localization and fast causal instance selection for multi-instance learning [J]. JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE), 2023, 58(9): 105-113, 126.
[2]	WANG Gang, XU Xin-shun*. A new Multi-instance learning method for scene classification [J]. J4, 2010, 45(7): 108-113.