MONITAUR

Wer sind wir? We are a research group focused on investigating security- and privacy-preserving solutions for Machine Learning (ML). With this project, we want to address the problem of violation of intellectual property (IP) rights in Internet services to make sharing e.g. ML solutions as-a-service more secure.

Für wen ist es? Our solution aims to protect complex program logic that can be considered intellectual property due to, for instance, their high-demanding development process. Hence, our target group are owners of such algorithms who want to share their solutions on the Internet (as an API, for instance) but avoid malicious exploitation and illegal copying of their intellectual property.

Was ist es? Making machine learning or other algorithmic solutions available on the Internet raises threats to their owners. A program logic may not be fully disclosed but is used within a service, which clients can use for sending requests and getting feedback. However, a malicious client can use that interaction channel to reveal the behaviour of the underlying algorithm and create an illegal copy. This way, a valuable model can be appropriated by parties who can illegitimately use or monetise it.

Wie funktioniert es? Program logic stealing can be mitigated by monitoring clients' behaviour and raising warnings if the behaviour looks suspicious. Hence, malicious clients are detected and suspended from using the service. Our main goal is to develop a toolkit with various monitoring techniques such that developers of software services can easily deploy them to protect the intellectual property of model owners.