SPEAKERS

"In recent years adversarial machine learning represents a growing cybersecurity topic since A.I. and machine learning applications are becoming more and more popular within all industrial sectors. In fact, as soon as a machine learning application is released in a production environment and exposed to a broader public, it could be targeted by adversarial machine learning attacks. Those attacks can have a variety of aims including tricking the models by providing deceptive input, model stealing, or sensitive information retrieval. Such attacks have been extensively explored in areas such as image classification and object detection, text generation, recommendation engines, sentiment analysis, credit scoring, spam detection, and email filtering.

In light of this, we believe it is now crucial to rethink traditional DevSecOps procedures when machine learning models are involved. In specific, different robustness and security checks must be introduced in order to avoid potential adversarial machine learning vulnerabilities.

This session will initially introduce the concept of adversarial machine learning attacks and relevant real-world cases. Subsequently, the most common attacks are explored including data poisoning, evasion attacks, and model extraction. Finally, a set of instruments and best practices are demonstrated in order to make sure that traditional DevSecOps pipelines are hardened against adversarial machine learning."