A Multiple Instance Learning Strategy for Combating Good Word Attacks on Spam Filters

By Zach Jorgensen, Yan Zhou, M. Inge. 2008

75 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Information Systems, Management of Technology and Innovation

How researchers see this paper

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

In Computer Networks and Communications

In Information Systems

In Management of Technology and Innovation

Citations in the corpus

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].”
Adversarial Feature Selection Against Evasion Attacks (2016, background): “It has been, therefore, widely adopted in previous work [10], [11], [15], [16], [23], [26], and in several real anti-spam filters, like SpamAssassin and SpamBayes.”
Multi-instance multi-label learning (2008, methodology): “Multi-instance learning techniques have already been applied to diverse applications including image categorization [17, 18], image retrieval [71, 84], text categorization [3, 60], web mining [86], spam detection [37], computer security [54], face detection [66, 76], computer-aided medical diagnosis”
Multiple Classifier Systems for Adversarial Classification Tasks (2009, background): “Most of the works proposed countermeasures against specific kinds of attacks for spam filtering and intrusion detection tasks (see for instance [1–3]), and only few of them proposed formal models of adversarial classification tasks [4, 5], or analysed the main issues raised by the application of mac”
One-and-a-Half-Class Multiple Classifier Systems for Secure Learning Against Evasion Attacks at Test Time (2015, background): “Due to the high dimensionality of the corresponding feature space, linear classifiers have been often considered [17,19,18] and implemented in real anti-spam filters, including e.”
Multiple classifier systems for robust classifier design in adversarial environments (2010, background): “In the spam filtering task, several works analysed the vulnerability of text classifiers against well known spammers’ tricks aimed at getting spam emails misclassified as legitimate [26,41,31,24,35].”
A brief introduction to weakly supervised learning (2018, methodology): “Multi-instance learning has been successfully applied to various tasks, such as image categorization/retrieval/annotation [20, 73, 90], text categorization [3, 66], spam detection [45], medical diagnosis [34], face/object detection [30, 76], object class discovery [101], object tracking [6], etc.”
Multiple instance learning: A survey of problem characteristics and applications (2016, background): “In [150], the content of emails is analyzed to detect spam.”
Evade Hard Multiple Classifier Systems (2009, background): “Actually, the issue of the hardness of evasion has been addressed only recently in the machine learning and pattern recognition literature, often with respect to specific tasks, classification systems and type of attacks [5, 7] (for instance, the so-called good words attack against the text classifi”
A review of machine learning approaches to Spam filtering (2009, background): “The importance of model update is highlighted by works studying the performance of filters (Jorgensen et al., 2008; Webb & Chitti, 2005) when dealing with disguised messages, where training using these messages improved the performance of the filters.”
A Survey on Machine Learning Techniques for Cyber Security in the Last Decade (2020, background): “[357] proposed a counterattack strategy using multiple instance learning.”