Intellectual Property Protection of Machine Learning Processes
IPP4ML will advance the research in the area of intellectual property (IP) protection, with the focus on the property that is a part of machine learning processes. It considers input data, especially innovative techniques for ownership protection of mixed-type databases, and techniques for emergent types of data, and their effects on the result of machine learning processes. Further, schemes for IP protection addressing machine learning models themselves will be investigated.
Digital data sharing is as old as digital data itself. Since data is a valuable asset to its owner, any type of unauthorised usage of shared data should be detected and sanctioned. With the advances in the area of Machine Learning (ML), outsourcing data and processing thereof became an increasingly popular trend among businesses. In this scope, the data is given to data management professionals that are involved in data mining, data classification etc., to make more use of the data. This can foster business growth by additional services (recommendation systems) or customer behaviour understanding. Furthermore, online services like Machine-Learning-as-a-Service (MLaaS) have an increasing popularity. Since available online, the ML models on such platforms became the target of the attackers who want to claim or use models as their own property.
The proposed research project seeks to advance the research in the area of IP protection, with the focus on the property that is a part of Machine Learning processes. Firstly, this includes the input data, which can be of various types (images, text, relational data, gene sequences, etc.). Novel aspects of research to be conducted include innovative techniques for ownership protection of mixed-type databases, applicable to real-world data bases and preserving coherence in the data. The project further addresses techniques of emergent types of data, and their effects on the result of Machine Learning processes. Secondly, novel schemes for IP protection addressing machine learning models themselves will be investigated, with the aim of protecting models from false ownership claim and model stealing attacks.
- This project is led by SBA Research.
- This FFG programme is sponsored by Nationalstiftung für Forschung, Technologie und Entwicklung and Österreich-Fonds. The focus lies on funding industrial PhD projects to improve qualifications of research and innovation staff in companies and non-university research institutions. An Industrial PhD project is performed by an employee of an Austrian company/non-university research institution, who is enrolled as a PhD student at a university during the whole project.
This project is funded by the FFG.