Two view learning: SVM-2K, Theory and Practice

By Jason D. R. Farquhar, David Roi Hardoon, Hongying Meng, John Shawe-Taylor, Sándor Szedmak. Neural Information Processing Systems, 2005

404 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/38a93d6d335ba266979534d7f8ee279f439b7af3

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Learning from Multiple Partially Observed Views - an Application to Multilingual Text Categorization (2009, background): “Our setting differs from previous multi-view learning studies [5] mainly on the straightforward generalization to more than two views a nd the use of view generating functions to induce the missing views from the observed ones.”
A Survey of Multi-View Representation Learning (2016, unknown): “[115] investigate the possibility of combining the two distinct stages of KCCA and SVM into a single optimization named as SVM-2K, which has been widely used in classification tasks where two views of the same phenomenon are available.”
A Graphbased Framework for Multi-Task Multi-View Learning (2011, methodology): “Therefore, we compare with the following algorithms: In our experiments, we apply SVM-2K on the multiple views of each task respectively; we apply SMTL and CASO on the common views of all the tasks.”
Improvability Through Semi-Supervised Learning: A Survey of Theoretical Results (2019, methodology): “Rosenberg and Bartlett (2007) and Farquhar et al. (2006) analyze a Rademacher complexity term under the multi-view assumption.”
Constrained Semi-supervised Learning in the Presence of Unanticipated Classes (2016, methodology): “Recent research on multiple kernel learning has proposed a number of approaches for combining kernels in the regularized risk minimization framework [38, 45, 78].”
Information Bottleneck Domain Adaptation with Privileged Information for Visual Recognition (2016, methodology): “We also train the SVM2k [14] and test only the SVM that uses the main view (indicated as SVM2k), and we perform kernel CCA (KCCA) [25] between main and auxiliary views, map the main view in feature space and train an SVM (indicated as KCCA).”
Semi-supervised learning by disagreement (2010, unknown): “Brefeld and Scheffer [16] extended SVM-2K [32], a supervised co-SVM that minimizes the training error as well as the disagreement between the two views, to semi-supervised learning and applied it to several tasks involving structured output variables such as multi-class classification, label sequenc”
Simple and Efficient Learning using Privileged Information (2016, methodology): “We compare our proposed new algorithm SVM2+ with the baseline algorithms, i.e. , SVM, SVM-2K [Farquhar et al. , 2005] and SVM1+.”
Recognizing RGB Images by Learning from RGB-D Data (2014, methodology): “Among these methods, SVM2K achieves the best result, as it can more effectively exploit depth information by learning two classifiers for both vi- 3The 10 common categories between the two datasets are calculator, cereal box, coffee mug, keyboard, flashlight, lightbulb, mushroom, ball, soda can, tom”
Generalized Multiview Analysis: A discriminative latent space (2012, background): “In [6], the authors obtained a multi-view version of SVM by constraining the one-dimensional outputs of individual SVM’s to be equal.”
A Correlation Approach for Automatic Image Annotation (2006, methodology): “In future work we would like to explore enhancing the annotation accuracy by combining several image descriptors[3] as well as examining a new non orthogonal representation of the keywords as labels for the MMR method.”
Classifier learning with hidden information (2015, methodology): “Multi-view learning tackles the problem where features from multiple sources (or views) are available during both training and testing [15, 16, 17].”
Bayesian exponential family projections for coupled data sources (2010, background): “, 2003) and discovering dependencies between images and associated text to be used as preprocessing for classification (Farquhar et al., 2006).”
A survey of multi-view machine learning (2013, methodology): “[16] introduced supervised SVM2K that was later extended to multi-view semi-supervised learning [46].”
One-Pass Learning with Incremental and Decremental Features (2016, background): “For example, in [11], [18], [19], they all required the consistency among different views.”