Support Vector Method for Novelty Detection

By Bernhard Schölkopf, Robert C. Williamson, Alex Smola, John Shawe-Taylor, John C. Platt. 1999

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

Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift (2018, methodology): “Popular approaches to anomaly detection include density estimation [6], margin-based approaches such as the one-class SVM [40], and the tree-based isolation forest method due to [30].”
Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty (2019, methodology): “Deep SVDD [Ruff et al., 2018] uses convolutional networks to extract features from the raw pixels all while modelling one class, like OC-SVMs.”
A systematic study of the class imbalance problem in convolutional neural networks (2017, methodology): “They include Support Vector Data Description (SVDD) method [67, 68], hyper-plane learning that separates outliers [69], neighbourhood based OSVM [70], Support Vector Mapping Convergence [71], or fuzzy OSVM [72, 73], just to mention some.”
Does Your Model Know the Digit 6 Is Not a Cat? A Less Biased Evaluation of "Outlier" Detectors (2018, background): “The popular one-class SVM [24] with a radial basis function learns a conservative region that covers the train data.”
Continual Learning in Practice (2019, background): “For the problem of detecting anomalies (abrupt but transient shifts), early approaches that scale well to higher dimensions include the one class Support Vector Machine (SVM) [25], which are ﬂexible but are computationally prohibitive.”
You Are What You Do: Hunting Stealthy Malware via Data Provenance Analysis (2020, background): “• One-Class SVM [93]: The algorithm trains a hyper-plane which separates all the training data from the origin while maximizing the distance from the origin to the hyper-plane.”
Scaling Out-of-Distribution Detection for Real-World Settings (2022, background): “Accordingly, research on out-of-distribution detection has a rich history spanning several decades (Schölkopf et al., 1999; Breunig et al., 2000; Emmott et al., 2015).”
Explainable AI: Interpreting, Explaining and Visualizing Deep Learning (2019, methodology): “The case for transparency has been made in many settings, including for government policy [69], business [44], charity [63], and algorithms [48].”
Extreme Value Theory for Open Set Classification - GPD and GEV Classifiers (2018, methodology): “rameters whose optimization is difﬁcult. The problem of distinguishing known and unknown objects is strongly related to the problem of estimating the support of the distribution of the known examples [6]. One direct way to perform open set recognition is to model the probability of an object of bein”
Latent Space Autoregression for Novelty Detection (2018, methodology): “Surprisingly, OC-SVM outperforms most deep learning based models in this setting.”
Attack Strength vs. Detectability Dilemma in Adversarial Machine Learning (2018, methodology): “To accomplish this, the optimization function differs from the traditional two-class SVM in that there is no opposing class data between which to draw an optimal hyperplane; instead, the goal is to maximize the distance between the classification hyperplane and the origin in the appropriate high-dim”
CutPaste: Self-Supervised Learning for Anomaly Detection and Localization (2021, background): “Anomaly detection under one-class classification setting, where we assume only the normal data is given during the training, has been widely studied [49, 56, 46, 63, 66, 13, 42, 36, 27].”
Support Vector Clustering (2002, methodology): “Some minor modifications are required to adapt it to the unsupervised training problem addressed here, see Schölkopf et al. (2000).”
Kernel PCA for novelty detection (2007, methodology): “Because of the success of SVM, attempts have been made to apply SVM to novelty detection [6–8].”
Uncertainty Estimation Using a Single Deep Deterministic Neural Network (2020, methodology): “Instead our approach is more related to pre-deep learning kernel methods (Quinn & Sugiyama, 2014; Sch¨olkopf et al., 2000), such as Gaussian processes which revert to a prior away from data, and Support Vector Machines, where the distance to the separating hyperplane is informative of the…”
Potential, challenges and future directions for deep learning in prognostics and health management applications (2020, background): “s labelling only a subset of the available data [149]. Fault detection can also be dened as a one-class classication problem [150]. However, dierently to previously proposed one-class formulations [151], the problem is dened in [150] as a regression problem where the healthy system conditions during”
Exploratory Learning (2013, background): “[24] defines a problem of learning a function over the data space that isolates outliers from class instances.”
Adversarially Learned Anomaly Detection (2018, methodology): “One Class Support Vector Machines (OC-SVM) [15] are a classic kernel method for novelty detection that learns a decision boundary around normal examples.”