Security Challenges in Information Systems

Scheduling:

05.05.17: 08:00 - 11:30 (Room 301)
Kickoff - Presentation of offered topics and students choose from these.

10.05.17: 10:00 - 11:30 (Room 348)
Tutorial for technical necessities (Latex, GIT, etc. ) and deadline for seminar registration

24.05.17: 11:30 - 13:00 (Room 348)
Proposal presentation. Each student presents his initial and future work (7 min + 3 min questions).

01.08.17: Submission of the slides for final presentations

08.08.17: 8.08.17 (Room 348) Final presentations

14.08.17: First draft of term paper

31.08.17 12:00 (commit to git repository) Final version of the term paper

Abstract:

Information systems have become the backbone of most organizations. These systems contain data of different degree of importance and confidentiality. Data confidentiality violations can have serious impact on business processes. Therefore, access control management to ensure that information and resources are available to only the authorized users has become more crucial. Orthogonal to this is that such data is usually analysed with machine learners that draw some rules from it. E.g. in fraud detection one might be interested in finding anomalies in the large quantities of recorded customer data via Machine Learning methods. However, adversarial learning teaches us that such learner are vulnerable to attacks. I.e., a crook aware of the fraud detection method might try to fool the machine learner such that the crook’s behaviour is declared non-anomalous. To avoid such detection faults or to be able make meaningful corrections to a model, it is significant to understand the reasoning of ML algorithms. This problem can be approached e.g. by building interpretable models or extracting the rules/boundaries from complex ‘black box’ algorithms.The goal of this seminar is to expose students to a wide range of research topics in the area of security in Information Systems. The seminar includes topics like access control models, adversarial learning, neural-based anomaly detection and algorithmic transparency.

Topics:

  • Classifier Evasion / Poisoning / G. Steinbuß

  • Location-Time Role Based Access Control Model / G. Suntaxi

  • Relation Based Access Control Model / G. Suntaxi
  • Enforcing Access Control Models on Outsourced Data / G. Suntaxi
  • Anomaly detection using Autoencoders /
  • Anomaly detection using Self-Organizing-Maps /
  • Interpretable models /
  • Model induction, model deconstruction /