Continuous Authentication

Background

In developing a continuous authentication system to overcome the limitations of password-based systems, each enrolled user has a series of typing vectors that can be regarded as a 'User Matrix'. This matrix consists of all typing vectors that can be definitely associated with the specific user and represents the vector space of what is currently known about the user's typing pattern (keystroke Biometric Capture). In order to match a users known typing vectors against unknown typing vectors (underlying process of continuous authentication), the Singular Value Decomposition of the User Matrix can be performed in order to define the eigenvectors associated with that matrix. Once these Eigenvectors are determined, then a series of vector transformations can be applied to them in order to generate any/all other vectors that a user could produce. The process of determining the type and sequence of these vector transformations is computationally difficult and requires the use of some efficient search technique or metaheuristic. The use of genetic algorithms to find the correct transformations and their proper sequence or to provide a useful 'rule' for determining them may show promise.

Project Description

Genetic Algorithms (GA) are ideally suited for applications in which the search space is very large or complex. Determining the proper type and sequence of vector transformations necessary to apply to an eigenvector, in order to reproduce a 'compatible vector' in a User's vector space, is one such problem. Successful implementation of GAs, to find these transformations or to allow for the devising of a metaheuristic to determine them empirically, would be a 'significant' step forward in building a reliable continuous authentication system. This project (known as Jellyfish) is a continuation of project RazorFish which focused on building a continuous authentication system based on eigenvector decomposition of a user's typing vectors. This project is designed to accomplish the following:

Project Deliverables

References

This exact work has not been previously explored but the following papers are related to the general process of using GA for this type of problem. (Represent work on same general principle).