Learning to classify with missing and corrupted features

By Ofer Dekel, Ohad Shamir, Lin Xiao. 2008

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 Artificial Intelligence

Citations in the corpus

Wild Patterns: Ten Years After the Rise of Adversarial Machine Learning (2017, positive): “Zero-sum games have been formulated to learn invariant transformations like feature insertion, deletion and rescaling [33– 35].”
Using Machine Teaching to Identify Optimal Training-Set Attacks on Machine Learners (2015, background): “Promising as the name suggests, we point out that robust learning is not an appropriate defense against training-set attacks.”
Evasion Attacks against Machine Learning at Test Time (2013, background): “Most of the previous work on evasion attacks assumes that the attacker can arbitrarily change every feature [8, 10, 12], but they constrain the degree of manipulation, e.”
Poisoning Attacks against Support Vector Machines (2012, methodology): “In response to the threat of adversarial data manipulation, several proposed learning methods explicitly account for certain types of corrupted data (Globerson & Roweis, 2006; Teo et al., 2008; Br¨uckner & Schef-fer, 2009; Dekel et al., 2010).”
Practical Evasion of a Learning-Based Classifier: A Case Study (2014, methodology): “Some methods of this kind have been recently proposed for learning on problems with potential feature deletion or corruption [43], [44].”
Analysis of classifiers’ robustness to adversarial perturbations (2015, background): “The security of machine learning algorithms at test time have also been previously examined in different scenarios, in particular when the adversary has only limited knowledge about the classifier (Biggio et al. 2013; Dekel et al. 2010; Srndic and Laskov 2014).”
Security Evaluation of Pattern Classifiers under Attack (2014, methodology): “Many authors implicitly performed security evaluation as a what-if analysis, based on empirical simulation methods; however, they mainly focused on a specific application, classifier and attack, and devised ad hoc security evaluation procedures based on the exploitation of problem knowledge and heur”
Security analysis of online centroid anomaly detection (2010, methodology): “A somewhat related approach has been developed for handling worst-case random noise; for example, random feature deletion (Globerson and Roweis, 2006; Dekel and Shamir, 2008).”
Towards Making Systems Forget with Machine Unlearning (2015, neutral): “[36] developed two techniques to make learning algorithms resilient against attacks.”
Adversarial Label Flips Attack on Support Vector Machines (2012, background): “Different types of feature noise have been extensively studied in several literature [4, 6, 9, 11].”
Vox Populi: Collecting High-Quality Labels from a Crowd (2009, background): “A somewhat related problem is designing machine learning algorithms that withstand spe-ciﬁc types of label-noise, either on the training set [KL93, Kea98], or on the test set [TGRS07, DS08].”
Online Anomaly Detection under Adversarial Impact (2010, background): “Some recent work in machine learning (Dalvi et al., 2004; Dekel and Shamir, 2008; Globerson and Roweis, 2006; Teo et al., 2008) and computer security (Barreno et al.”
Static prediction games for adversarial learning problems (2012, background): “, 2009) or performs an arbitrary feature transformation (Teo et al., 2007; Dekel and Shamir, 2008; Dekel et al., 2010) have been studied.”
Optimization Models for Machine Learning: A Survey (2019, background): “From a mathematical perspective, such security issues have been formerly expressed via min-max approaches where the learner’s and the attacker’s loss functions are antagonistic [81, 104, 146].”
Supervised classification and mathematical optimization (2013, unknown): “A different perspective is presented in [76].”
Machine learning in adversarial environments (2010, methodology): “Dekel et al. (2010) apply machine learning techniques to create two-class linear classifiers designed to perform well when an adversary can corrupt or delete a constrained number of input features.”
Feature Weighting for Improved Classifier Robustness (2009, methodology): “We also note that in [6], the authors proposed a similar method (FMICO) which allows one to assign importance to features a priori to countermeasure feature over or undertraining.”
Learning with Marginalized Corrupted Features (2013, methodology): “Learning with MCF is quite different from previous approaches for this setting (Dekel & Shamir, 2008; Globerson & Roweis, 2006): it does not learn under a worst-case scenario, but the (arguably) more common average-case scenario by considering all possible corrupted observations.”
Dropout: a simple way to prevent neural networks from overfitting (2014, background): “Previous work of Globerson and Roweis (2006); Dekel et al. (2010) explored an alternate setting where the loss is minimized when an adversary gets to pick which units to drop.”