Classical data management systems mainly try to improve the efficiency and scalability of storing and processing data. Our projects in the area of Trusted Data Management introduce a new perspective: A data management systems should provide users with additional trust guarantees that increase the confidence of the user while interacting with the system itself and with other users through the system.
Such guarantees include the security to prevent unauthorized access and manipulation of the stored data by a user, or even the administrator. Moreover, the explainability of results is an important aspect, since it increases the confidence of the user that the system indeed returned a correct and integer (query) result.
In the following, you can find a selection of active and past projects at our lab in this exciting research area:
Decentralized Federated ML
Decentralized Federated Machine Learning (ML) aims at enabling a collaborative machine learning setup in which parties that do not trust each other can work together to learn powerful ML models. Decentralized Federated ML uses Blockchain technology to decentralize previously centralized components in a ML setup and introduces novel protocols that prevent participants from manipulating the learning procedure or the resulting model.
Privacy-related Operations in Smart Storage
It is important to ensure that personally identifiable information (PII) is protected within large distributed systems and that it is used only for its intended purposes. We believe that future storage solutions should include, in addition to emerging compute offload, also privacy-related operators -- these have to be provided, however, without slowing down the overall system! We are working on various topics in this area, including privacy-preserving pre-processing inside Smart Storage nodes, and efficient GDPR-compliant Key-Value Stores built using FPGAs.