The scientific supervisor of SibIASA took part in the webinar “Künstliche Intelligenz datenschutzkonform entwickeln” (“Developing artificial intelligence in compliance with data protection rules”), held by the XING Ambassador Community with the support of DATAKONTEXT GmbH.
Ideas for using differential privacy machine learning techniques to create Trustworthy AI were discussed. Development work by the European company DQ0 (https://dq0.io) was referred to as an example of the given approach being implemented.
At SibIASA, an approach is being developed that allows the confidentiality of data to be maintained when using artificial intelligence thanks to the automated design of machine learning technologies by self-adaptive stochastic optimization algorithms directly for the user’s tasks. In this case, the user receives an artificial intelligence system for his/her task without the need to involve the developer of such a system and disclose confidential data to him/her.
More information on the topic
The speaker provided links for more detailed information on the topic:
- The Algorithmic Foundations of Differential Privacy: https://www.cis.upenn.edu/~aaroth/Papers/privacybook.pdf
- Differential Privacy: A Primer for a Non-technical Adience: https://privacytools.seas.harvard.edu/files/privacytools/files/pedagogical-document-dp_0.pdf
- DQ0 Software-Plattform für vertauenswürdige KI und Differential Privacy: https://dq0.io/de/
- OpenDP: https://opendifferentialprivacy.github.io/
- OpenMinded: https://www.openmined.org/
- Tensorflow Privacy: https://github.com/tensorflow/privacy
- EU Ethics guidelines for trustworthy AI: https://ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai