Transactions on Machine Learning and Data Mining (ISSN: 1865-6781)
Volume 4 - Number 1 - July 2011 - Pages 30-52
Moving Targets in Computer Security and Multimedia Retrieval
Dip. Ing. Elettrica ed Elettronica, Università di Cagliari, Italy
The Internet era is changing the way Pattern Recognition has been defined in the past years. New applications are emerging whose characteristics can be hardly matched against the typical problem setting. The typical formulation of a pattern recognition problem assumes that data can be subdivided into a number of classes on the basis of the values of a set of suitable features. Supervised techniques assume that data classes are given in advance, and the goal is to find the most suitable set of features and classification algorithm that allows the effective partition of data. On the other hand, unsupervised techniques allow discovering the “natural” data classes in which data can be partitioned, for a given set of features. These approaches are showing their limitations to handle the challenges issued by applications where the definition of data classes is not uniquely fixed. As a consequence, the tasks of feature definition, and classifier training should be adapted to this changing environment. Two applications from different domains share similar characteristics in this respect, namely, Intrusion Detection in computer systems and Multimedia Retrieval. In intrusion detection, the adversary can carefully craft attack patterns so that they are undetected by the employed detector. On the other hand, the retrieval of multimedia data by content is biased by the high subjectivity of the concept of similarity. In this paper, the issues of the two application scenarios will be discussed, and some effective solutions and future research directions will be outlined.
Download Paper (519 KB)