About the Project
Machine learning is extremely pervasive today with variety of applications, some of which use privacy sensitive user data. In general, computing parties (e.g., cloud providers) and data owners (e.g., users) are not the same, and this setting introduces multiple layer of threats. The existing literature on privacy-preserving machine learning mostly addresses three main objectives (in isolation): privacy of the data used for learning a model or as input to an existing model, privacy of the model, or privacy of the model’s output. In addition, most of the existing techniques are not applicable to real-world applications, as they require heavy pre-processing and communicating large amounts of input data from the data owners to the computing parties, which might not be possible in some Internet-of-Things (IoT) devices that have limited memory and computational capabilities.
This project will perform research and development of practical privacy-preserving machine learning technologies to address the challenges faced in real-world applications. More specifically, the student will study advanced secure computation technologies such as differential privacy, homomorphic encryption and secure multiparty computations, and evaluate challenges in these technologies in terms of their applicability to machine learning technologies. A special attention will be given to practical challenges and restrictions (e.g., memory and computational capabilities of the data generators - IoT devices) that arise in applying these technologies to real-world applications.
In addition, the PhD student will be supervised jointly by research experts in two world-leading institutions – the University of Manchester (UoM) and the Institute for Infocomm Research (I²R) Singapore. The student will be hosted by both organisations: Year 1 & 4 at UoM in the UK and Year 2 & 3 at I²R in Singapore.
For informal enquiries about the project, please contact Dr Mustafa A. Mustafa: [Email Address Removed]
Entry Requirements:
Applications should be submitted online and candidates should make direct contact with the Manchester supervisor to discuss their application directly. Applicants must have obtained, or be about to obtain, at least an upper second class honours degree (or equivalent) in a relevant subject.
Funding Notes
This project is available to UK/EU candidates. Funding covers fees (UK/EU rate) and stipend for four years. Overseas candidates can apply providing they can pay the difference in fees and are from an eligible country. Candidates will be required to split their time between Manchester and Singapore, as outlined on www.manchester.ac.uk/singaporeastar.
As an equal opportunities institution we welcome applicants from all sections of the community regardless of gender, ethnicity, disability, sexual orientation and transgender status. All appointments are made on merit.
References
[1] Ahmad Al Badawi, Bharadwaj Veeravalli, Chan Fook Mun, Khin Mi Mi Aung: High-Performance FV Somewhat Homomorphic Encryption on GPUs: An Implementation using CUDA. IACR Trans. Cryptogr. Hardw. Embed. Syst. 2018(2): 70-95 (2018)
[2] Ahmad Al Badawi, Yuriy Polyakov, Khin Mi Mi Aung, Bharadwaj Veeravalli, Kurt Rohloff: Implementation and Performance Evaluation of RNS Variants of the BFV Homomorphic Encryption Scheme. IACR Cryptology ePrint Archive 2018: 589 (2018)
[3] Ahmad Al Badawi, Jin Chao, Jie Lin, Chan Fook Mun, Sim Jun Jie, Benjamin Hong Meng Tan, Xiao Nan, Khin Mi Mi Aung, Vijay Ramaseshan Chandrasekhar: The AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data with GPUs. IACR Cryptology ePrint Archive 2018: 1056 (2018)
[4] Iraklis Symeonidis, Abdelrahaman Aly, Mustafa A. Mustafa, Bart Mennink, Siemen Dhooghe, Bart Preneel: SePCAR: A Secure and Privacy-Enhancing Protocol for Car Access Provision. In 22nd European Symposium on Research in Computer Security (ESORICS 2017), LNCS 10493, S. N. Foley, D. Gollmann, and E. Snekkenes (eds.), Springer-Verlag, pp. 475-493, 2017.
[5] Konstantinos Sechidis, Gavin Brown: Simple strategies for semi-supervised feature selection. Machine Learning Journal. Machine Learning, 107(2), 357-395, 2018.