Integrating constraints and metric learning in semi-supervised clustering

By Mikhail Bilenko, Sugato Basu, Raymond J. Mooney. 2004

987 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Epidemiology, Signal Processing

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

In Epidemiology

In Signal Processing

Citations in the corpus

Learning to cluster in order to Transfer across domains and tasks (2017, methodology): “The experiments in table 1 show a clear trend that utilizing the pairwise constraints jointly for both metric learning and minimizing the clustering loss achieves the best performance, including both MPCK-means and CCN.”
A Survey on Metric Learning for Feature Vectors and Structured Data (2013, background): “ion brought by the set of unlabeled pairs, i.e., pairs of training examples that do not belong to the sets of positive and negative pairs: U = {(xi,xj) : i6= j,(xi,xj) ∈ S ∪D}/ . An early approach by Bilenko et al. (2004) combined semi-supervised clustering with metric learning. In the following, we”
Recent Advances in Open Set Recognition: A Survey (2018, background): “PENDIGITS [136]: has a total of 10992 samples from 10 classes, where each class has around 1099 samples with 16 features.”
A survey of content-based image retrieval with high-level semantics (2007, methodology): “[80], in order to automatically annotate database images for retrieval purpose, the system firstly cluster image regions into region clusters using a variant of k-means clustering called pair-wise constraints k-means (PCK-means) [91].”
Semi-supervised Kernel Metric Learning Using Relative Comparisons (2016, methodology): “The MPCK-Means algorithm by Bilenko et al. [2004] is one of the ﬁrst to combine metric learning with semi-supervised clustering and ML/CL constraints.”
Neural network-based clustering using pairwise constraints (2015, methodology): “Thus, the scheme is not the same as the semi-supervised clustering framework in previous works (Wagstaff et al., 2001; Bilenko et al., 2004; De Amorim & Mirkin, 2012) where their unlabeled data contribute to calculating the centers of the clusters.”
Metric Learning: A Survey (2013, methodology): “[3], which also employs Mahalanobis metric learning for clustering.”
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.”
A Semantic Loss Function for Deep Learning with Symbolic Knowledge (2017, background): “To sustain this progress, and alleviate the need for more labeled data, there is a growing interest into utilizing unlabeled data to augment the predictive power of classiﬁers (Stewart & Ermon, 2017; Bilenko et al., 2004).”
Effective Unconstrained Face Recognition by Combining Multiple Descriptors and Learned Background Statistics (2011, neutral): “Alternatively, a number of authors have proposed tailoring similarity measures to available training data by applying learning techniques (e.g., [28], [29], [30], [31], [32]).”
Learning Concept Embeddings with Combined Human-Machine Expertise (2015, methodology): “We compare against other semi-supervised learning and constrained clustering systems: Label Propagation [4], the multiclass version of the Constrained Spectral Clustering KMeans (CSPKmeans) method described in [36], and Metric PairwiseConstrained KMeans (MPCKmeans) [5].”
Deep Amortized Clustering (2019, positive): “(Cai et al., 2009) 0.372 0.498 LSC (Chen & Cai, 2011) 0.376 0.500 CSP (Wang et al., 2014) 0.812 0.812 MPCK-means (Bilenko et al., 2004) 0.871 0.816 KCL (Hsu et al., 2017) 0.889 0.874 MCL (Hsu et al., 2019) 0.897 0.893 While the metrics for DAC were computed after at most 100 forward passes,…”
Multi-class Classification without Multi-class Labels (2019, methodology): “…et al., 2014) 62.4% 46.9% 0.770 0.781 SKLR (Amid et al., 2016) 66.9% 46.8% 0.791 0.760 CSP (Wang et al., 2014) 62.5% 65.4% 0.812 0.812 MPCK-means (Bilenko et al., 2004) We use constrained clustering algorithms as the baselines since they can use the pairwise inputs from a similarity prediction…”
Hierarchical Density Estimates for Data Clustering, Visualization, and Outlier Detection (2015, methodology): “During the label expansion they use a method to change the distance between objects that resembles distance learning [Bilenko et al. 2004].”