Privacy and Machine Learning – Pitfalls and Solutions | Lecture with Frank Kargl

When processing personally identifiable information (PII) in machine learning systems, there are a number of potential pitfalls by which data may be abused and user’s privacy may be negatively affected. Beyond, there are also security and ethical considerations that should be considered. The talk will illustrate and categorize those challenges with use-case examples from the mobility domain like detection of aggressive driving behavior through isolation forests or privacy-preserving geofencing through decision trees. Prof. Kargl will also present recent research results that show how to mitigate this seeming contradiction between services based on personal data and protection of privacy through modern privacy-enhancing technologies (PETs). Through clever technological solutions, one can analyze and use PII without compromising personal privacy in inappropriate ways. In some cases, this even simplifies obligations from the GDPR as the system may not even access personal data. The talk will also briefly address how a privacy engineering process can be integrated into system and software engineering.