MLC with Missing Labels(MLML):在多标签问题中,标签很可能会丢失。例如,是的XML问题是,标记者不可能遍历所有的标签,所以标记者通常只给出一个子集,而不是所有的监督信息。文献中解决这个问题的技术主要是基于图片和标签空间(或Latent标签空间)Low-Rank的方法、基于概率图模型的方法。
Deep Embedding Methods:早期的Embedding方法通常使用线性投影,将PCA、Compressed Sensing等方法引入多标签学习问题。一个很自然的问题是,线性投影真的能够很好地挖掘标签之间的相关关系吗?同时,在SLEEC[3]的工作中也发现某些数据集并不符合Low-Rank假设。因此,在2017年的工作C2AE[7]中,Yeh等将Auto-Encoder引入了多标签学习中。由于其简单易懂的架构,很快有许多工作Follow了该方法,如DBPC[8]等。
Fig. 4. The architecture of Canonical-Correlated Autoencoder (C2AE).C2AE learns a latent space L via NN mappings of Fx, Fe, and Fd. X and Y are the instance and label matrices respectively.
Deep Learning for Challenging MLC:深度神经网络强大的拟合能力使我们能够有效地处理更多更困难的工作。因此我们发现近年的趋势是在CV、NLP和ML几大Community,基本都会有不同的关注点,引入DNN解决MLC的问题,并根据各自的问题发展出自己的一条线。
DL for MLC with unseen labels:这一领域的发展令人兴奋,今年ICML的工作DSLL[12]探索了流标签学习,也有许多工作[13]将Zero-Shot Learning的架构引入MLC。
Advanced Deep Learning for MLC:有几个方向的工作同样值得一提。首先是CNN-RNN[14]架构的工作,近年有一个趋势是探索Orderfree的解码器[15]。除此之外,爆火的图神经网络GNN同样被引入MLC,ML-GCN[16]也是备受关注。特别的,SSGRL[17]是我比较喜欢的一篇工作,结合了Attention机制和GNN,motivation比较强,效果也很不错。
讲了这么多方法论,但追溯其本源,这么多纷繁复杂的问题依然是由任务驱动的,正是有许许多多现实世界的应用,要求我们设计不同的模型来解决尺度更大、监督更弱、效果更强、速度更快、理论性质更强的MLC模型。因此,在文章的最后一部分,我们介绍了近年多标签领域一些最新的应用,如Video Annotation、Green Computing and 5G Applications、User Profiling等。在CV方向,一个趋势是大家开始探索多标签领域在视频中的应用[24]。在DM领域,用户画像受到更多关注,在我们今年的工作CMLP[25]中(下图),就探索了对刷单用户进行多种刷单行为的分析。不过,在NLP领域,似乎大家还是主要以文本分类为主,XML-Repo[2]中的应用还有较多探索的空间,所以我们没有花额外的笔墨。
Figure 6: Some services that a malicious service platform provides. The dishonest merchants can freely select different combinations of these services, e.g. Two-day Task.
1. Chang W C, Yu H F, Zhong K, et al. Taming Pretrained Transformers for Extreme Multi-label Text Classification\[C\]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining. 2020: 3163-3171.
3. Bhatia K, Jain H, Kar P, et al. Sparse local embeddings for extreme multi-label classification\[C\]//Advances in neural information processing systems. 2015: 730-738.
4. Chu H M, Yeh C K, Frank Wang Y C. Deep generative models for weakly-supervised multi-label classification\[C\]//Proceedings of the European Conference on Computer Vision \(ECCV\). 2018: 400-415.
5. Zhang M L, Fang J P. Partial multi-label learning via credible label elicitation\[J\]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.
6. Wang H, Liu W, Zhao Y, et al. Discriminative and Correlative Partial Multi-Label Learning\[C\]//IJCAI. 2019: 3691-3697.
7. C. Yeh, W. Wu, W. Ko, and Y. F. Wang, “Learning deep latent space for multi-label classification,” in AAAI, 2017, pp. 2838–2844.
8. X. Shen, W. Liu, Y. Luo, Y. Ong, and I. W. Tsang, “Deep discrete prototype multilabel learning,” in IJCAI, 2018, pp. 2675–2681.
9. You R, Zhang Z, Wang Z, et al. Attentionxml: Label tree-based attention-aware deep model for high-performance extreme multi-label text classification\[C\]//Advances in Neural Information Processing Systems. 2019: 5820-5830.
10. Chang W C, Yu H F, Zhong K, et al. Taming Pretrained Transformers for Extreme Multi-label Text Classification\[C\]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining. 2020: 3163-3171.
11. Durand T, Mehrasa N, Mori G. Learning a deep convnet for multi-label classification with partial labels\[C\]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 647-657.
12. Z. Wang, L. Liu, and D. Tao, “Deep streaming label learning,” in ICML, 2020.
13. C. Lee, W. Fang, C. Yeh, and Y. F. Wang, “Multi-label zero-shot learning with structured knowledge graphs,” in CVPR, 2018, pp. 1576–1585.
14. Wang J, Yang Y, Mao J, et al. Cnn-rnn: A unified framework for multi-label image classification\[C\]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 2285-2294.
15. Yazici V O, Gonzalez-Garcia A, Ramisa A, et al. Orderless Recurrent Models for Multi-label Classification\[C\]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020: 13440-13449.
16. Chen Z M, Wei X S, Wang P, et al. Multi-label image recognition with graph convolutional networks\[C\]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 5177-5186.
17. T. Chen, M. Xu, X. Hui, H. Wu, and L. Lin, “Learning semanticspecific graph representation for multi-label image recognition,” in ICCV, 2019, pp. 522–531.
18. M. J. Er, R. Venkatesan, and N. Wang, “An online universal classifier for binary, multi-class and multi-label classification,” in IEEE International Conference on Systems, Man, and Cybernetics, 2016, pp. 3701–3706.
19. H. Chu, K. Huang, and H. Lin, “Dynamic principal projection for cost-sensitive online multi-label classification,” Machine Learning, vol. 108, no. 8-9, pp. 1193–1230, 2019.
20. S. Boulbazine, G. Cabanes, B. Matei, and Y. Bennani, “Online semi-supervised growing neural gas for multi-label data classification,” in IJCNN, 2018, pp. 1–8.
21. H. Yu, P. Jain, P. Kar, and I. S. Dhillon, “Large-scale multilabel learning with missing labels,” in Proceedings of the 31th International Conference on Machine Learning, ICML 2014, Beijing, China, 21-26 June 2014, 2014, pp. 593–601.
22. W. Gao and Z. Zhou, “On the consistency of multi-label learning,” Artificial Intelligence, vol. 199-200, pp. 22–44, 2013.
23. W. Liu and X. Shen, “Sparse extreme multi-label learning with oracle property,” in ICML, 2019, pp. 4032–4041.
24. X. Zhang, H. Shi, C. Li, and P. Li, “Multi-instance multi-label action recognition and localization based on spatio-temporal pretrimming for untrimmed videos,” in AAAI. AAAI Press, 2020, pp. 12 886–12 893.
25. H. Wang, Z. Li, J. Huang, P. Hui, W. Liu, T. Hu, and G. Chen, “Collaboration based multi-label propagation for fraud detection,” in IJCAI, 2020.