Semi-Supervised Classification by Low Density Separation

By Olivier Chapelle, Alexander Zien. 2005

1,040 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

Deep learning via semi-supervised embedding (2008, methodology): “The first three are small scale datasets used in the same experimental setup as found in [9, 24, 10].”
Class-Incremental Domain Adaptation (2020, background): “Effectively, the total loss Ls = Ls1 +Ls2 enforces the Cluster Assumption at the U-space (via Ls1) that enhances the model’s generalizability [6,14], and, mitigates the overconfidence issue (via Ls2) thereby reducing the discriminative bias towards the source domain.”
Smooth Neighbors on Teacher Graphs for Semi-Supervised Learning (2017, background): “It has a special case named cluster or low density separation assumption, which states that the decision boundary should lie in low density regions, not crossing high density regions [9].”
Curriculum Labeling: Revisiting Pseudo-Labeling for Semi-Supervised Learning (2020, background): “When training a classiﬁer, a common assumption is that the decision boundary should lie in low-density regions in order to improve generalization (Chapelle and Zien 2005); therefore unlabeled samples that lie either near or far from labeled samples should be more informative for decision boundary…”
Deep Unsupervised Domain Adaptation: A Review of Recent Advances and Perspectives (2022, background): “Several UDA approaches based on a popular clustering assumption (Chapelle and Zien, 2005) are proposed in the context of semi-supervised training, which indicates target domain samples from the same class are likely to be distributed closely as a cluster.”
Towards Making Unlabeled Data Never Hurt (2011, methodology): “It was discovered that S3VMs realize the low-density assumption [12] which states that the decision boundary will go across low-density regions.”
Beyond the point cloud: from transductive to semi-supervised learning (2005, methodology): “The experimental protocol is based on (Joachims, 2003) for the WebKB dataset and (Chapelle & Zien, 2005) for other datasets.”
Semi-supervised time series classification (2006, background): “Since finding the exact TSVM solution is NP-hard, several approximation algorithms have been proposed [4][10][15][16][21].”
Semi-Automated Data Labeling (2020, methodology): “Relying on the cluster assumption Chapelle and Zien (2005) our bootstrapping algorithm partitions the data using K-means clustering.”
Optimization Techniques for Semi-Supervised Support Vector Machines (2008, methodology): “In Section 4 we review this class of methods (Chapelle and Zien, 2005; Chapelle et al., 2006a; Fung and Mangasarian, 2001; Collobert et al., 2006).”
Semisupervised Neural Networks for Efficient Hyperspectral Image Classification (2010, background): “, the ∇TSVM, still has a cubic cost and requires storing a l + u sample kernel matrix, which precludes its use in remote sensing [28].”
Discriminative Semi-Supervised Feature Selection Via Manifold Regularization (2009, methodology): “The representative approaches in this category include tranductive support vector machine (SVM) and semi-supervised SVM [11], [12], [27], [54], [59], [60].”
Large Scale Transductive SVMs (2006, methodology): “In fact, Chapelle and Zien (2005) show that, as these methods provide a new representation, one can just as easily run TSVM on the new representation.”