* Hassan Khan,Aaron Atwater,and Urs Hengartner

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

Why Implicit Authentication （ IA ）

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

Implicit Authentication  As a second line of defense  Measurement patterns of device use that are gathered from the device users without requiring deliberate actions.  When an adversary bypass a pass-lock,IA may be able to detect it

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

some IA system  TAP :a virtual key typing based authentication system for mobile device,transparently enhance the secrity of the mobile device in two stage:login stage and post- login stage  continuous mobile authentication using virtual key typing biometrics  Contemporary IA:device-centric  Application-centric implicit authentication ： enable fine grained authentication control with high accuracy  Towards Application-Centric Implicit Authentication on Smartphones

Implicit Authentication at platform level  OS or app framework is responsible for providing IA to all apps in an app independent manner;  Limitation: 1.different apps have different characteristics 2.Performance and battery life 3.The difference in app characteristics reduce the accuracy of IA 4.not be able to distinguish which activities within an app require protection 5.need to be managed by the platform developers or some central authority.

Solution for Limitation  These limitations can be circumvented by enabling the apps themselves to implicity authenticate users.  Delegate IA to apps:usability and flexibility  App developer requires domain-specific knowledge and significant effort to provide IA support: IA library

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

Challenges  Marking the resulting library easy to use for app developers  Be flexible enough to cater for app-specific functionality  Be extensible by the researchers developing novel IA schemes  Shouldn’t have a noticeable impact on performance or battery life to end users

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

Itus  A framework that separates the domain knowledge of IA from its deployment  App developers can provide IA in their apps easily  Allowing IA developers to iteratively contribute improvement to the framework.  Implement as a library at the user level  Itus Oracle:a tool for app developer

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

For app developer Easy to use

For app developer configurable

For app developer customized

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

For researchers and IA developers

Prefabs  Prefabs are pre-configured instances of the Itus Agent  …can specify classifiers,feature subsets,parameters,etc.  …allow an app developer to pick a prefab and go

Machine Learning Toolkit  Contains Classifier objects, each implementing a single machine learning algorithm  Developers implement only train and classify methods in order to add new algorithms

Event Processor  D evelopers create Measurement classes (e.g. Touch, Keystroke, Movement)  User input events are centrally managed and delivered by a Dispatcher object  Measurement objects only need to process input data into a numeric feature vector

Outline Background Introduction for IA Previous work on IA challenge Introduction to Itus Itus for app developer Itus for IA developer Performance & conclusion

Performance  Created three Itus Prefabs using different machine learning algorithms and feature sets  Touch-based, movement-based, and keystroke-based  Added to two open-source Android apps  Zirco web browser and TextSecure SMS messenger  Measured library overhead in terms of CPU, memory, and battery

CPU Overhead

Memory Overhead

Conclusion Itus is an application-level implicit authentication framework for Android that is: easy can be deployed by developers with minimal LoC extensible subcomponents can be improved iteratively Flexible support for application-specific functionality fast minimal, tuneable performance overhead