Learning a Distance Metric from Relative Comparisons

By Matthew Schultz, Thorsten Joachims. 2003

784 citations recorded by the source corpus.

Subfields: Representation & Self-Supervised Learning

How researchers see this paper

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

In Representation & Self-Supervised Learning

Citations in the corpus

Invariant Information Clustering for Unsupervised Image Classification and Segmentation (2018, background): “The latter includes objectives such as generative autoencoder image reconstruction [48], triplets [46] and spatial-temporal order or context prediction [37, 12, 17], for example predicting patch proximity [30], solving jigsaw puzzles [41] and inpainting [43].”
Learning Spread-Out Local Feature Descriptors (2017, background): “One commonly used triplet loss is the ranking loss [18]:”
Sampling Matters in Deep Embedding Learning (2017, methodology): “For example, the triplet loss [26, 36] is one of the most prominent methods that emerged from convex optimization.”
No Fuss Distance Metric Learning Using Proxies (2017, methodology): “To address some of the issues in this and similar work [13] a Semi-Hard negative mining approach was introduced in [12].”
A Survey on Metric Learning for Feature Vectors and Structured Data (2013, methodology): “ce induced by a kernel function and thereby combine the best of both worlds, in the spirit of what is done in SVM. Some metric learning approaches have been shown to be kernelizable (see for instance Schultz and Joachims, 2003; Shalev-Shwartz et al., 2004; Hoi et al., 2006; Torresani and Lee, 2006; ”
Decomposition-Based Transfer Distance Metric Learning for Image Classification (2014, methodology): “A popular categorization of the DML methods is: supervised DML [1], [3], [9], [10], [12], [18] and unsupervised DML [19], [20], according to the underlying learning paradigm.”
Learning To Classify Images Without Labels (2020, neutral): “and testing. Dataset CIFAR10 CIFAR100-20 STL10 Metric ACC NMI ARI ACC NMI ARI ACC NMI ARI K-means [50] 22.9 8.7 4.9 13.0 8.4 2.8 19.2 12.5 6.1 SC [55] 24.7 10.3 8.5 13.6 9.0 2.2 15.9 9.8 4.8 Triplets [42] 20.5 { { 9.94 { { 24.4 { { JULE [54] 27.2 19.2 13.8 13.7 10.3 3.3 27.7 18.2 16.4 AEVB [26] 29.1”
A Modulation Module for Multi-task Learning with Applications in Image Retrieval (2018, background): “Another important related research area is metric learning [30,31,32,33], which mostly focuses on designing an optimization objective to find a metric to maximize the inter-class distance while minimizing the intra-class distance.”
Metric Learning: A Survey (2013, methodology): “At a high level, the method of Schultz and Joachims [61] may be viewed as employing a squared Frobenius norm regularizer with relative distance constraints, under the additional assumption that the matrix learned is a diagonal matrix.”
Deep Feature Learning for EEG Recordings (2015, neutral): “As introduced by Schultz & Joachims (2004), a relative similarity constraint ( a, b, c ) describes a relative comparison of the trials a , b and c in the form “ a is more similar to b than a is to c .””
Robustness and generalization for metric learning (2012, methodology): “where k M k = k M F in [1] and k M k = k M k 1 , 2 in [8].”
An incremental face recognition system based on deep learning (2017, methodology): “As to loss function, we employ triplet loss [5]to obtain the more precise features.”