ML- and IT Security

Ansprechpartner: Nicolas Müller - Fraunhofer AISEC

nicolas.mueller@aisec.fraunhofer.de

Geplante Vorlesungszeiten: Montag, 10.00 Uhr bis 12.00 Uhr im Seminarraum 00.13.009A

                                                  Donnerstag, 10.00 Uhr bis 12.00 Uhr im Hörsaal 00.04.011

This lecture provides an overview over the application of machine learning (ML) to IT-Security, and security-aspects of machine-learning.

 Content:

• Overview of the application areas 'ML for IT security' and 'IT security for ML'.

• Fundamentals and advanced concepts for anomaly detection, including both semi- and unsupervised algorithms

• Basics of dimension reduction for data visualization and preprocessing

• Fundamentals of Natural Language Processing (NLP) for IT Security: Spam and Malware Detection using Learning Algorithms

• Basic and advanced concepts of adversarial machine learning: attackability of systems using white/grey/black box attacks and corresponding defenses

• Possibilities and limitations of transferability of such attacks from the lab to the real world

• Basic concepts of both creation and detection of audio deepfakes (spoofing) incl. extraction of corresponding features (such as MFCC)

IN0001 and one of the following is recommended:

·       IN2332, Statistical Modeling and Machine Learning

·       IN2346, Introduction to Deep Learning

After successful completion of this module, participants will have an overview of three areas in the intersection of IT security and machine learning: 

1.     Machine learning for IT security, i.e., the application of ML for enhancing security in information systems. This includes anomaly detection algorithms for detecting e.g. network attacks, but also basic approaches for spam and malware detection. Classical methods for detecting attacks, spam or malware (e.g. signature-based algorithms) are not covered. 

2.     IT Security for ML: Students will know basic and advanced attacks on ML systems under different threat models and will be able to describe and classify their advantages and disadvantages. They can describe defenses and their limitations and evaluate their effectiveness. 

3.     Deepfake / Audio Spoofing: Participants will know the basic concepts in audio deepfake / audio spoofing. This includes both the creation and detection of audio deepfakes using ML. Students will be familiar with the most basic signal processing techniques as they pertain to feature extraction.