Jelle Vos

Daniël Vos, Jelle Vos, Tianyu Li et al.

Decision trees are interpretable models that are well-suited to non-linear learning problems. Much work has been done on extending decision tree learning algorithms with differential privacy, a system that guarantees the privacy of samples within the training data. However, current state-of-the-art algorithms for this purpose sacrifice much utility for a small privacy benefit. These solutions create random decision nodes that reduce decision tree accuracy or spend an excessive share of the privacy budget on labeling leaves. Moreover, many works do not support continuous features or leak information about them. We propose a new method called PrivaTree based on private histograms that chooses good splits while consuming a small privacy budget. The resulting trees provide a significantly better privacy-utility trade-off and accept mixed numerical and categorical data without leaking information about numerical features. Finally, while it is notoriously hard to give robustness guarantees against data poisoning attacks, we demonstrate bounds for the expected accuracy and success rates of backdoor attacks against differentially-private learners. By leveraging the better privacy-utility trade-off of PrivaTree we are able to train decision trees with significantly better robustness against backdoor attacks compared to regular decision trees and with meaningful theoretical guarantees.

Daniël Vos, Leon Overweel, Wouter Raateland et al.

Millions of shipping containers filled with goods move around the world every day. Before such a container may enter a trade bloc, the customs agency of the goods' destination country must ensure that it does not contain illegal or mislabeled goods. Due to the high volume of containers, customs agencies make a selection of containers to audit through a risk analysis procedure. Customs agencies perform risk analysis using data sourced from a centralized system that is potentially vulnerable to manipulation and malpractice. Therefore we propose an alternative: DEFenD, a decentralized system that stores data about goods and containers in a secure and privacy-preserving manner. In our system, economic operators make claims to the network about goods they insert into or remove from containers, and encrypt these claims so that they can only be read by the destination country's customs agency. Economic operators also make unencrypted claims about containers with which they interact. Unencrypted claims can be validated by the entire network of customs agencies. Our key contribution is a data partitioning scheme and several protocols that enable such a system to utilize blockchain and its powerful validation principle, while also preserving the privacy of the involved economic operators. Using our protocol, customs agencies can improve their risk analysis and economic operators can get through customs with less delay. We also present a reference implementation built with Hyperledger Fabric and analyze to what extent our implementation meets the requirements in terms of privacy-preservation, security, scalability, and decentralization.