Online and batch learning of pseudo-metrics

By Shai Shalev-Shwartz, Yoram Singer, Andrew Y. Ng. International Conference on Machine Learning, 2004

342 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/af74e4ceebfd8a9bf2e3e52ab1d262fdb2653bc4

How researchers see this paper

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

In Artificial Intelligence

Citations in the corpus

Robustness and generalization for metric learning (2012, methodology): “Online methods [2, 5, 9] nevertheless oﬀer guarantees, but only in the form of regret bounds assessing the deviation between the cumulative loss suﬀered by the online algorithm and the loss induced by the best hypothesis that can be chosen in hindsight.”
Distance Metric Learning for Large Margin Nearest Neighbor Classification (2005, background): “In fact, as shown by many researchers [2] [6] [7] [8] [12] [13], kNN classification can be significantly improved by learning a distance metric from labeled examples.”
Learning Globally-Consistent Local Distance Functions for Shape-Based Image Retrieval and Classification (2007, background): “An advantage of the dual solver is that, like online learning methods such as [20], it finds a near-optimal solution very quickly.”
Learning a Mahalanobis distance metric for data clustering and classification (2008, background): “Furthermore, some online distance metrics learning algorithms [39,47] have been proposed recently for the situations where the data points are collected sequentially.”
Online Passive-Aggressive Algorithms (2003, methodology): “…conducted further research on applications of the PA algorithmic framework for learning margin-based sufﬁx trees (Dekel et al., 2004b), pseudo-metrics (Shalev-Shwartz et al., 2004b), hierarchical classiﬁcation (Dekel et al., 2004a), and segmentation of sequences (Shalev-Shwartz et al., 2004a).”
Regularized Distance Metric Learning: Theory and Algorithm (2009, background): “Note that compared to the online metric learning algorithm in [9], the proposed online learning algorithm for metric learning is advantageous in that (i) it is computationally more efficient by avoiding projecting a matrix into a SDP cone, and (ii) it has a provable regret bound while [9] only prese”
Learning Multi-View Neighborhood Preserving Projections (2011, background): “Relatively little prior work, however, exists regarding the integration of multi-view data, and the methods that do exists are often tailored with specific application domains in mind: Vinokourov et al. (2002) and Hardoon et al.”
Survey on distance metric learning and dimensionality reduction in data mining (2014, background): “Pseudo-metric online learning (POLA) (Shalev-Shwartz et al. 2004) falls into the category of supervised metric learning with pairwise constraints.”
A Survey on Metric Learning for Feature Vectors and Structured Data (2013, background): “The first projection is different from POLA (since the generalized cosine has a normalization factor that depends on M) but the authors manage to derive a closed-form solution.”
Randomized Clustering Forests for Image Classification (2008, background): “This function is generally inspired by the Mahalanobis distance dðx; yÞ 1⁄4 ðx yÞAðx yÞ, as in [35] or, more recently, [58], [22], [50], [3], [21], [56].”
Distance Metric Learning with Eigenvalue Optimization (2012, methodology): “Indeed, Srebro et al. (2007) proposed to employ standard SDP solvers (e.”
Latent Coincidence Analysis: A Hidden Variable Model for Distance Metric Learning (2012, methodology): “To handle larger data sets, we plan to explore online strategies for distance metric learning [18], possibly based on Bayesian [19] or conﬁdence-weighted updates [20].”
Randomized Smoothing for Stochastic Optimization (2011, background): “Another example is the problem of metric learning [37, 33], in which the learner is given a set of n points { a 1 , . . . , a n } ⊂ R d and a matrix B ∈ R n × n indicating which points are close together in an unknown metric.”
Improving Semantic Embedding Consistency by Metric Learning for Zero-Shot Classiffication (2016, methodology): “To solve the optimization problem, we do not follow the approach proposed in [43] since we also learn the attribute embedding part WX jointly with the metric embedding WA.”
Communication/Computation Tradeoffs in Consensus-Based Distributed Optimization (2012, methodology): “Following the formulation in [14], we have a data set { u j , v j , s j } mj =1 with u j , v j ∈ R d and s j = {− 1 , 1 } signifying whether or not u j is similar to v j (e.g., similar if they are from the same class).”