Robust Support Vector Machine Training via Convex Outlier Ablation

By Linli Xu, Koby Crammer, Dale Schuurmans. 2006

177 citations recorded by the source corpus.

How researchers see this paper

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

In Optimization & Training

Citations in the corpus

Towards the Science of Security and Privacy in Machine Learning (2016, background): “Their approach does not impact the convexity of the optimization problem unlike previous attempts [78], [79], which reduces the impact of the defense mechanism on performance.”
Learning kernel logistic regression in the presence of class label noise (2014, methodology): “Comparison of the different kernel set size at 30% level, averaged over 20 random runs. comparison with SVM serves the purpose of a baseline, i.e. to see how far would the gold-standard off-the-shelf method allow one to treat label noise as a normal part of any classification problem.”
SoK: Security and Privacy in Machine Learning (2018, background): “Their approach does not impact the convexity of the optimization problem unlike previous attempts [78], [79], which reduces the impact of the defense mechanism on performance.”
Robust kernel density estimation (2008, methodology): “Robust surrogate losses for kernel-based classifiers have also been studied (Xu et al., 2006).”
Trading convexity for scalability (2006, background): “More recently Xu et al. (2006) proposed a variant of SVMs where the Hinge Loss is also replaced by a non-convex loss.”
Support Vector Machines with the Ramp Loss and the Hard Margin Loss (2011, methodology): “[40] solve a semidefinite programming relaxation of SVM with the ramp loss, but the procedure is computationally intensive for as few as 50 observations.”
Multiple Kernel Learning from Noisy Labels by Stochastic Programming (2012, methodology): “Note that an alternative approach is to consider the best case analysis (a strategy taken in robust SVM (Xu et al., 2006; Yu et al., 2011)) by minimizing the robust hinge loss, which can be defined as minp∈Q ∑ i(piℓi+1−pi), where ℓi denotes the loss on ith example, to address the uncertainty…”
Enhancing one-class support vector machines for unsupervised anomaly detection (2013, background): “The above formulation will be relaxed similar to what was proposed in the original work [33] into a semi-definite problem.”
Support Vector Machines Under Adversarial Label Noise (2011, background): “…+ C n∑ i=1 ξi s. t. yi(w >xi + b) ≥ 1− ξi, i = 1, . . . , n, ξi ≥ 0, i = 1, . . . , n. (2) where the margin is maximized by minimizing 12w >w, and the variables ξi, i = 1, . . . , n (referred to as slack variables) represent the extent to which the samples xi, i = 1, . . . , n, violate the margin.”
Robust twin support vector machine for pattern classification (2013, unknown): “1⁄2tn2 ð g n 2Þ AL (3), 1⁄2ln2 w n b AL nþ1: ð75Þ According to Lemma 15 of [25], we can obtain”
Robust support vector machines based on the rescaled hinge loss function (2017, background): “[13] proposed semi-definite programming along with convex relaxation.”
Noise Tolerance Under Risk Minimization (2011, background): “for noisy points [5] and restricting the effects of noisy poi nts on the classifier [6], [7] are some of the well known tricks to tackle the label noise.”