Design of robust classifiers for adversarial environments

By Battista Biggio, Giorgio Fumera, Fabio Roli. 2011

39 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Safety, Robustness & Interpretability

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 Computer Networks and Communications

In Safety, Robustness & Interpretability

Citations in the corpus

Evasion Attacks against Machine Learning at Test Time (2013, background): “Alternatively, one may explicitly model the attack distribution, as in [4]; or add the generated attack samples to the training set.”
Security Evaluation of Pattern Classifiers under Attack (2014, methodology): “learning algorithms robust to a limited amount of data contamination [14], [29], and classification algorithms robust to specific exploratory attacks [6], [23], [26], [32].”
Statistical Meta-Analysis of Presentation Attacks for Secure Multibiometric Systems (2016, methodology): “In [13], [37] spoofing-aware score-level fusion rules were proposed, as variants of the well-known LLR rule [38]:”
Security Evaluation of Support Vector Machines in Adversarial Environments (2014, background): “Generative classifiers can be modified, by explicitly modeling the attack distribution, as in [11], and discriminative classifiers can be modified similarly by adding generated attack samples to the training set.”
Is data clustering in adversarial settings secure? (2013, background): “Other approaches may more directly encode explicit assumptions on how the data distribution changes under attack, similarly to [5].”
Support vector machines under adversarial label contamination (2015, methodology): “, using game-theoretical models for classification [15, 20–22], probabilistic models of the data distribution drift under attack [23, 24], and even multiple classifier systems [25–27].”
Cracking Classifiers for Evasion: A Case Study on the Google's Phishing Pages Filter (2016, background): “Similarly, for approaches based on probabilistic models [16][48], the adversary can also easily construct a hard-to-predict attack instance based on cracked knowledge.”
Pattern Recognition Systems under Attack: Design Issues and Research Challenges (2014, background): “To learn secure generative classifiers, instead, the distribution of the manipulated attack samples should be explicitly modeled, as proposed in [19, 64].”
One-and-a-Half-Class Multiple Classifier Systems for Secure Learning Against Evasion Attacks at Test Time (2015, background): “Therefore, it does not make any significant difference in terms of risk if regions of the feature space which are not densely populated by training data are classified as legitimate or malicious [9,10,11].”