Metric Learning by Collapsing Classes

By Amir Globerson, Sam T. Roweis. 2005

770 citations recorded by the source corpus.

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

Generative Continual Concept Learning (2019, background): “, when wrong concepts (or classes) across two tasks are matched in the embedding space (Globerson and Roweis 2006).”
A Survey on Metric Learning for Feature Vectors and Structured Data (2013, methodology): “MCML (Globerson & Roweis) Shortly after Goldberger et al., Globerson and Roweis (2005) proposed MCML (Maximally Collapsing Metric Learning), an alternative convex formulation based on minimizing a KL divergence between pij and an ideal distribution, 16.”
Metric Learning with Adaptive Density Discrimination (2015, background): “In this case, a ubiquitous solution is to enforce semantic similarity: examples of each class are demanded to be tightly clustered together, far from examples of other classes (for example, Schroff et al. (2015); Norouzi et al. (2012); Globerson & Roweis (2006); Chopra et al. (2005)).”
Dimensionality Reduction of Multimodal Labeled Data by Local Fisher Discriminant Analysis (2007, methodology): “Globerson and Roweis (2006) proposed using the following alternate iterative procedure for obtaining U MCML .”
Metric Learning: A Survey (2013, methodology): “As follow-up work to NCA, Globerson and Roweis [25] proposed a variant to NCA which has the desirable property of convexity.”
Unbiased Metric Learning: On the Utilization of Multiple Datasets and Web Images for Softening Bias (2013, methodology): “0 20 40 60 80 100 0.24 0.26 0.28 0.3 0.32 0.34 0.36 0.38 0.4 Average Results # of training examples per class A cu rr ac y UML +Trseen+Vaweb +Teunseen LMNN +Trseen+Vaseen+Teunseen mtLMNN +Trseen+Vaseen+Teunseen MCML +Trseen+Vaseen+Teunseen SML +Trseen+Vaseen+Teunseen 0 20 40 60 80 100 0.2 0.25 0.3 0”
Unsupervised Domain Adaptation with Similarity Learning (2017, methodology): “[60, 17, 18, 59, 40]) used a learned similarity measure together with nearest neighbors classifier.”
Person re-identification by probabilistic relative distance comparison (2011, methodology): “Related work–Although it has not been exploited for person re-identification, distance learning is a well-studied problem with a large number of methods reported in the literature [16, 5, 17, 7, 17, 12, 15, 9, 1].”
PCCA: A new approach for distance learning from sparse pairwise constraints (2012, methodology): “For PRDC and MCC, we reproduce the figures given in [33].”
Distance-Based Image Classification: Generalizing to New Classes at Near-Zero Cost (2013, unknown): “Other closely related methods are metric learning by collapsing classes [26] and neighborhood component analysis [27].”
Linear dimensionality reduction: survey, insights, and generalizations (2014, methodology): “…relevant component analysis for DML (Bar-Hillel et al., 2003), neighborhood component analysis (Goldberger et al., 2004), collapsing classes (Globerson and Roweis, 2005), discriminative component analysis (Peltonen et al., 2007), latent coincidence analysis (Der and Saul, 2012), and an…”
Learning Deep Representation for Imbalanced Classification (2016, background): “For instance, Globerson and Roweis [14] learn the Mahalanobis distance which requires all the same-class samples collapse to one point.”
Toward a Quantitative Survey of Dimension Reduction Techniques (2019, methodology): “…that represents sample similarity by proximity; Interactive Document Maps ( IDMAP ) [75], which maps data by a fast projection, then reﬁne the result using a force scheme; and Maximally Collapsing Metric Learning ( MCML ) [76], that use convex optimization to learn a quadratic Gaussian metric.”
Local Descriptors Encoded by Fisher Vectors for Person Re-identification (2012, background): “PCCA [20] reports state-of-the-art results for person re-identiﬁcation, improving over Maximally Collapsing Classes [27], ITML [18] or LMNN-R [28].”
Metric Learning to Rank (2010, methodology): “For comparison purposes, we also trained metrics with Large Margin Nearest Neighbor (LMNN) (Weinberger et al., 2006), Neighborhood Components Analysis (NCA) (Goldberger et al., 2005), and Metric Learning by Collapsing Classes (MLCC) (Globerson & Roweis, 2006).”
Central Subspace Dimensionality Reduction Using Covariance Operators (2011, unknown): “This class-labeled supervised dimension reduction framework, starting from the well-known Fisher Discriminant Analysis and generalized in the form of metric learning [11], [12], [29], [32], has received considerable attention.”
Learning a Nonlinear Embedding by Preserving Class Neighbourhood Structure (2007, background): “As noted by [8] learning a similarity measure is closely related to the problem of feature extraction, since for any ﬁxed (cid:1) , any feature extraction algorithm can be thought of as learning a similarity metric.”
Supervised principal component analysis: Visualization, classification and regression on subspaces and submanifolds (2011, methodology): “This body of research includes various approaches such as classical Fisher’s Discriminant Analysis (FDA) [3], the large family of methods known as Metric Learning (ML) [4–12], the family of Sufficient Dimension Reduction (SDR) algorithms [13–20], and the method proposed by Bair et al.”