I am a Research Fellow in the Software Innovation Institute at the Australian National University. My primary area of research is preserving the privacy of individuals through different techniques including distributed machine learning, differential privacy, and federated learning. I am also interested in the area of privacy-preserving record linkage (PPRL) especially in developing different privacy attacks on PPRL techniques.
Currently, I am working under the supervision of Prof. Graham Williams.
My research interests generally lie in the areas of data privacy and security, data mining, privacy-preserving record linkage, distributed machine learning, differential privacy, and federated learning.
Potential student research projects
1. Privacy-preserving record linkage in the presence of missing data
Privacy-preserving record linkage (PPRL) is the process of identifying records that refer to the same real-world entity across data sources held by different organizations while preserving the privacy of the entities whose records are being linked. These data sources often contain sensitive information of entities which needs to be protected. Because of the lack of unique entity identifiers across the data sources, the linking is usually based on quasi-identifying attributes such as names and addresses.
Handling missing values is one of the major challenges in different processing domains. In PPRL it is no different. Because of the missing values that occur in quasi-identifying attributes, linking records across databases can result in lower linkage quality.
The proposed project aims to investigate the means of overcoming the problem of missing values in PPRL to achieve high linkage quality. It explores the idea of using similar records in a data source to try and estimates the similarities of records that have missing values in them. The project will also investigate how different encoding techniques can be used to calculate attribute and relational similarities between records in the presence of missing values to identify matches and non-matches between different data sources in a privacy-preserving way.
2. Privacy-preserving active learning for data linkage
Data linkage is the process of identifying matching data points (or records) from multiple different data sources, which is required in a variety of applications, for example, cybersecurity and health analytics applications. However, the machine learning algorithms for data linkage significantly dependent on domain expertise for manual labelling of records for accurate linkage because the training data is generally not available. In real-world applications has several constraints which include cost-constraints, privacy-constraints, and fairness-constraints. These constraints could lead to poor linkage quality and privacy issues.
In this project, we aim to study privacy-preserving active learning subject to the above three constraints. Privacy-preserving active learning techniques should not only reduce the cost of labelling by selecting informative data samples to be labelled but also need to reduce the privacy risk of re-identification. The project has the following objectives.
- Compare existing algorithms for selecting data samples for manual labelling.
- Develop novel active learning algorithms using privacy-enhancing technologies such as Differential privacy mechanisms.
- Design fairness-aware algorithms for privacy-preserving active learning.
- Evaluate (both theoretically and empirically) the trade-off between privacy, the accuracy of linkage, fairness of linkage, and cost provided by the designed algorithms.
3. Privacy-preserving knowledge graph merging/linking and querying
Knowledge Graph (KG) has been popularly constructed and used by more and more organisations due to its ability to connect different types of data in meaningful ways and support rich data services. A KG is a heterogeneous graph composed of entities (nodes) and relations (edges), and in some KGs there are also properties (features) and labels for entities. However, the capabilities of KGs are limited by the data within organisations as there are certain privacy and security concerns of integrating KGs across organisations.
Within this project, we aim to explore how the KGs across organisations can be integrated and utilised using privacy enhancing technologies, such as differential privacy. We consider two important issues in KG isolation, which are
- Privacy-preserving knowledge graph merging – addresses the problem of identifying common entities in across KGs using perturbation techniques while balancing the privacy and utility trade-off.
- Privacy-preserving knowledge graph query - addresses the problem of querying over securely stored graph databases. In the presence of multiple private knowledge graphs from different organisations it would be challenging to query the graphs to find matching entities across graphs while protecting the privacy of the matched entities. The project aims to solve this challenge.
If you are interested in any of these projects, contact me via email@example.com
- Anushka Vidanage, Thilina Ranbaduge, Peter Christen, Rainer Schnell. 2020. “A Graph Matching Attack on Privacy-Preserving Record Linkage”. In ACM Conference on Information and Knowledge Management (CIKM), Ireland, 1485–1494.
- Anushka Vidanage, Thilina Ranbaduge, Peter Christen, Sean Randall. 2020. “A Privacy Attack on Multiple Dynamic Match-key based Privacy-Preserving Record Linkage”. In International Journal of Population Data Science (IJPDS), 5 (1).
- Anushka Vidanage, Thilina Ranbaduge, Peter Christen, Rainer Schnell. 2019. "Efficient Pattern Mining Based Cryptanalysis for Privacy-Preserving Record Linkage". In IEEE International Conference on Data Engineering (ICDE), Macau, 1698-1701.
- Thilina Ranbaduge, Anushka Vidanage, Sirintra Vaiwsri, Rainer Schnell, Peter Christen. 2018. “Evaluating hardening techniques against cryptanalysis attacks on Bloom filter encodings for record linkage”. In International Journal of Population Data Science (IJPDS), 3 (4).
- Peter Christen, Anushka Vidanage, Thilina Ranbaduge, Rainer Schnell. 2018. “Pattern mining based cryptanalysis of Bloom filters for privacy-preserving record linkage”. In Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD), Melbourne, 530-542.
- Anushka Vidanage, Isuru Dharmadasa, Buddika Jayasinghe, Chamath Keppitiyagama. 2016. "Reusing discarded-smartphone capabilities on quadcopters: The rationale, benefits and issues". In International Conference on Advances in ICT for Emerging Regions (ICTer), Negombo, 229-236.
- Australia–Germany joint research cooperation scheme fund, 2021 and 2022.
German academic exchange service (DAAD)
- SIGIR student travel grant, (2020)
Conference on Information and Knowledge Management (CIIKM)
The Visualise Your Thesis (VYT) competition finalist, (2020) [See presentation here]The Australian National University
3MT People's Choice Award and College Level Winner (2019) [See presentation here]Three Minute Thesis competitionThe Australian National University
ANU Postgraduate Research Scholarship (2017)The Australian National University
ANU HDR Merit Scholarship (2017)The Australian National University
Professor V. K. Samaranayake Medal for the Most Outstanding Graduate of UCSC (2015)University of Colombo School of Computing
Best Final Year Project Award for the 4th year research project (group) (2015)University of Colombo School of Computing
Bachelor of Science in Information and Communication Technology (Honours) - 1st ClassUniversity of Colombo School of Computing (UCSC), Sri Lanka - http://ucsc.cmb.ac.lk2011 - 2016
LecturerThe Australian National UniversityPresent - 2021
Tutor (Casual Sessional Academic)The Australian National University2021 - 2018
Research OfficerThe Australian National UniversityPresent - 2019
Teaching InstructorUniversity of Colombo School of Computing (UCSC)2016 - 2017
Entrepreneur, Software and Web DeveloperThink Abstract Solutions (PVT) Ltd.2014 - 2017