ABSTRACT
In data mining applications such as crowdsourcing and privacy-preserving data mining, one may wish to obtain consolidated predictions out of multiple models without access to features of the data. Besides, multiple models usually carry complementary predictive information, model combination can potentially provide more robust and accurate predictions by correcting independent errors from individual models. Various methods have been proposed to combine predictions such that the final predictions are maximally agreed upon by multiple base models. Though this maximum consensus principle has been shown to be successful, simply maximizing consensus can lead to less discriminative predictions and overfit the inevitable noise due to imperfect base models. We argue that proper regularization for model combination approaches is needed to alleviate such overfitting effect. Specifically, we analyze the hypothesis spaces of several model combination methods and identify the trade-off between model consensus and generalization ability. We propose a novel model called Regularized Consensus Maximization (RCM), which is formulated as an optimization problem to combine the maximum consensus and large margin principles. We theoretically show that RCM has a smaller upper bound on generalization error compared to the version without regularization. Experiments show that the proposed algorithm outperforms a wide spectrum of state-of-the-art model combination methods on 11 tasks.
Supplemental Material
- Acharya Ayan, Hruschka Eduardo, R., Ghosh Joydeep, Sarwar Badrul, and Ruvini Jean-David. Probabilistic combination of classifier and cluster ensembles for non-transductive learning. In SDM, 2013.Google Scholar
- Acharya Ayan, R. Hruschka Eduardo, Ghosh Joydeep, and Acharyya Sreangsu. An optimization framework for semi-supervised and transfer learning using multiple classifiers and clusterers. In CoRR, 2012.Google Scholar
- P. L. Bartlett. The sample complexity of pattern classification with neural networks: The size of the weights is more important than the size of the network. IEEE Trans. Inf. Theor., 2006. Google ScholarDigital Library
- Asa Ben-Hur, David Horn, Hava T. Siegelmann, and Vladimir Vapnik. Support vector clustering. Journal of Machine Learning Research, 2002. Google ScholarDigital Library
- Stephen Boyd and Lieven Vandenberghe. Convex Optimization. Cambridge University Press, 2004. Google ScholarDigital Library
- Ralph Allan Bradley. Rank analysis of incomplete block designs: Ii. additional tables for the method of paired comparisons. Biometrika, 41(3/4):pp. 502--537, 1954.Google ScholarCross Ref
- Ralph Allan Bradley and Milton E. Terry. Rank analysis of incomplete block designs: I. the method of paired comparisons. Biometrika, 39(3/4):pp. 324--345, 1952.Google ScholarCross Ref
- Sébastien Bubeck and Ulrike von Luxburg. Nearest neighbor clustering: A baseline method for consistent clustering with arbitrary objective functions. Journal of Machine Learning Research, 2009. Google ScholarDigital Library
- Xi Chen, Paul N. Bennett, Kevyn Collins-Thompson, and Eric Horvitz. Pairwise ranking aggregation in a crowdsourced setting. WSDM, 2013. Google ScholarDigital Library
- Janez Dem\vsar. Statistical comparisons of classifiers over multiple data sets. Journal of Machine Learning Research, 2006. Google ScholarDigital Library
- Inderjit S. Dhillon, Yuqiang Guan, and Brian Kulis. Kernel k-means: spectral clustering and normalized cuts. In SIGKDD, 2004. Google ScholarDigital Library
- John Duchi, Shai Shalev-Shwartz, Yoram Singer, and Tushar Chandra. Efficient projections onto the l1-ball for learning in high dimensions. ICML, 2008.Google ScholarDigital Library
- Wang Fei, Wang Xin, and Li Tao. Generalized cluster aggregation. In IJCAI, 2009. Google ScholarDigital Library
- Xiaoli Zhang Fern and Carla E. Brodley. Solving cluster ensemble problems by bipartite graph partitioning. ICML, 2004.Google ScholarDigital Library
- Jing Gao, Wei Fan, Deepak Turaga, Olivier Verscheure, Xiaoqiao Meng, Lu Su, and Jiawei Han. Consensus extraction from heterogeneous detectors to improve performance over network traffic anomaly detection. In INFOCOM, 2011.Google ScholarCross Ref
- Jing Gao, Feng Liang, Wei Fan, Yizhou Sun, and Jiawei Han. Graph-based consensus maximization among multiple supervised and unsupervised models. In NIPS, 2009.Google Scholar
- Ralf Herbrich, Tom Minka, and Thore Graepel. Trueskill(tm): A bayesian skill rating system. NIPS.Google Scholar
- Tao Li and Chris Ding. Weighted Consensus Clustering. SDM, 2008.Google ScholarCross Ref
- Tao Li, Chris Ding, and Michael I. Jordan. Solving consensus and semi-supervised clustering problems using nonnegative matrix factorization. ICDM, 2007. Google ScholarDigital Library
- Xudong Ma, Ping Luo, Fuzhen Zhuang, Qing He, Zhongzhi Shi, and Zhiyong Shen. Combining supervised and unsupervised models via unconstrained probabilistic embedding. IJCAI, 2011. Google ScholarDigital Library
- Meila Marina and Shortreed Susan. Regularized spectral learning. Journal of Machine Learning Research, 2006.Google Scholar
- Kaixiang Mo, Erheng Zhong, and Qiang Yang. Cross-task crowdsourcing. KDD, 2013. Google ScholarDigital Library
- Andrew Y. Ng, Michael I. Jordan, and Yair Weiss. On spectral clustering: Analysis and an algorithm. In NIPS, 2001.Google ScholarDigital Library
- H. Paugam-Moisy, A. Elisseeff, and Y. Guermeur. Generalization performance of multiclass discriminant models. In Neural Networks, 2000. IJCNN 2000, 2000. Google ScholarDigital Library
- Arun Rajkumar and Shivani Agarwal. A statistical convergence perspective of algorithms for rank aggregation from pairwise data. In ICML, 2014.Google Scholar
- Rion Snow, Brendan O'Connor, Daniel Jurafsky, and Andrew Y. Ng. Cheap and fast--but is it good?: Evaluating non-expert annotations for natural language tasks. EMNLP, 2008. Google ScholarDigital Library
- Alexander Strehl and Joydeep Ghosh. Cluster ensembles - a knowledge reuse framework for combining multiple partitions. Journal of Machine Learning Research, 2003. Google ScholarDigital Library
- Ben Taskar, Carlos Guestrin, and Daphne Koller. Max-margin markov networks. In NIPS. 2004.Google Scholar
- Vladimir Vapnik. Statistical learning theory. Wiley, 1998.Google ScholarDigital Library
- Hongjun Wang, Hanhuai Shan, and Arindam Banerjee. Bayesian cluster ensembles. In SDM, 2009.Google ScholarCross Ref
- Naiyan Wang and Dit-Yan Yeung. Ensemble-based tracking: Aggregating crowdsourced structured time series data. In ICML, 2014.Google Scholar
- Pu Wang, Carlotta Domeniconi, and Kathryn Blackmond Laskey. Nonparametric bayesian clustering ensembles. In ECML PKDD, 2010. Google ScholarDigital Library
- Ruby C. Weng and Chih-Jen Lin. A bayesian approximation method for online ranking. J. Mach. Learn. Res., 12, 2011. Google ScholarDigital Library
- Sihong Xie, Wei Fan, and Philip S. Yu. An iterative and re-weighting framework for rejection and uncertainty resolution in crowdsourcing. In SDM, 2012.Google ScholarCross Ref
- Sihong Xie, Xiangnan Kong, Jing Gao, Wei Fan, and Philip S. Yu. Multilabel consensus classification. In ICDM, 2013.Google ScholarCross Ref
- Yan Yan, Romer Rosales, Glenn Fung, and Jennifer Dy. Active learning from crowds. ICML, 2011.Google Scholar
- Jinfeng Yi, Tianbao Yang, Rong Jin, A.K. Jain, and M. Mahdavi. Robust ensemble clustering by matrix completion. ICDM, 2012. Google ScholarDigital Library
- Yi Zhang and Jeff Schneider. Maximum margin output coding. ICML, 2012.Google Scholar
- Jun Zhu, Amr Ahmed, and Eric P. Xing. Medlda: maximum margin supervised topic models for regression and classification. ICML, 2009. Google ScholarDigital Library
Index Terms
- Class-distribution regularized consensus maximization for alleviating overfitting in model combination
Recommendations
Heterogeneous source consensus learning via decision propagation and negotiation
KDD '09: Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data miningNowadays, enormous amounts of data are continuously generated not only in massive scale, but also from different, sometimes conflicting, views. Therefore, it is important to consolidate different concepts for intelligent decision making. For example, to ...
A Novel Regularized Model for Third-Order Tensor Completion
Inspired by the accuracy and efficiency of the <inline-formula><tex-math notation="LaTeX">$\gamma$</tex-math></inline-formula>-norm of a matrix, which is closer to the original rank minimization problem than nuclear norm minimization (NNM), we generalize ...
Loose L1/2 regularised sparse representation for face recognition
Sparse representation (or sparse coding) has been applied to deal with frontal face recognition. Two representative methods are the sparse representation‐based classification (SRC) and the collaborative representation‐based classification (CRC), in which ...
Comments