My third year research project was titled ‘Privacy‑Preserving Neural Network Inference via Homomorphic Encryption’. 

Machine Learning as a Service is becoming widespread in many critical fields. This presents a problem as a third party must store and process sensitive data. Privacy Preserving Machine Machine Learning proposes using homomorphic encryption to solve this security risk. This project aimed to reproduce a state-of-the-art research paper, which introduced a ResNet-20 model under a bootstrapped CKKS scheme, to analyse how practical this promising research is. Findings showed that although significant advancements have been made, computational cost and complexity mean that this technology is still far from a mature solution to the important question of how to ensure privacy of data in machine learning.