情報処理学会 第80回全国大会講演要旨

This paper proposes a privacy preserving machine learning (ML) prediction protocol for intelligence application on mobile terminals. Since the mobile terminals have very limited computational resources, client-server implementation are widely used for the mobile application. But this approach may have information leakage and privacy issue. The server or third person can see and collect the user's privacy sensitive data easily. Our proposed protocol can be useful to solve these problems. The protocol provides dividing neural network on the mobile devices and the server which has only black box calculator. The experimental results reveal the protocol works while keeping prediction performance compared with the original ML model.