Description
Privacy-preserving record linkag(PPRL) aims to identify records referring to the same real-world entity across different databases without revealing sensitive identifiers. For example, a hospital and an insurance provider may wish to determine whether “John Smith, Sydney” in one database and “Jon Smith, Sidney” in another refer to the same person, without exposing their full records. This is challenging because real-world data are often noisy, incomplete, and inconsistent, with spelling variations, typographical errors, and unstructured entries. Classical PPRL methods based on Bloom filters, secure computation, and differential privacy have made significant progress, but important limitations remain. Bloom-filter encodings can leak structural information, fuzzy-matching increases computational and communication costs, and differentially private PPRL methods often suffer from reduced linkage accuracy due to noise injection. In addition, quantum attacks threaten several classical communication and cryptographic schemes, motivating privacy-preserving protocols whose security can rely on quantum information principles. At the same time, quantum private set-intersection protocols, provide us significantly more security than their classical counterparts. However, the use of quantum methods for fuzzy record linkage remains largely unexplored. This project proposes a framework for quantum privacy-preserving record linkage(QPPRL). We develop and compare two complementary quantum-assisted linkage mechanisms. The first is a Bloom-filter-based protocol in which records are converted into q-grams, encoded as Bloom filters, represented through quantum Bloom-filter states, and compared using quantum overlap estimation to obtain Dice-style fuzzy similarity scores. The second uses hybrid quantum neural networks for entity matching, where quantum feature maps and variational circuits learn record-pair similarity functions. Differential privacy is incorporated into the learning-based approach to control information leakage during training and inference. The project compares these QPPRL approaches in terms of linkage accuracy, privacy leakage, communication cost, computational complexity, and robustness to fuzzy variations, establishing QPPRL as a distinct direction beyond quantum private set intersection.
| I am the presenting author | Yes |
|---|