Privacy Preserving Multi-Agent Planning
Cloud and distributed computing became a hot topic in recent years as gradually larger amounts of data are processed digitally, off site, and in distributed manner. A typical cloud solution is provided by large external companies such as Google, Microsoft, or Amazon. Provided that we need to process sensitive information in such cloud solutions, secure computation is the only viable way. There are many motivational examples. Medical data and personal information has to be treated as confidential by the law. Other type of confidential data are industrial secrets or corporate know-how. Probably the most sensitive data are of defense and military forces. When the sensitive data describe processes, which need to be coordinated among more parties, we face a problem of how to preserve privacy during collaborative planning. For instance, if different hospitals need to plan sequences of medical procedures for their patients using a 3rd party computational cluster, they need to coordinate the procedures based on the private medical information of their patients. Especially in the context of Industry 4.0, if several companies collaboratively produce goods, they need to keep their knowhow about their internal production processes private. To carry out their cooperative production, they need coordinated plans how to produce the goods together. Their planning process, however, is not allowed to leak any
private knowhow of the in-house production processes. Finally, in the example of military coalition operations the armies have to be coordinated, however as the coalition partners are from different countries, their private information has to be kept secret.
In this project are working on the description, formalization, and analysis of privacy preservation in multi-agent planning. We are evaluating existing multi-agent planning algorithms and planning tasks from perspective of privacy preservation. We aim to develop algorithms both for estimating and preventing privacy leakage during privacy preserving multi-agent planning. The grand challenge and the key objective of this project is a privacy-preserving multi-agent planner with parameterizable trade-off between privacy preservation, completeness, and efficiency.