Hardware-Accelerated Mode-Switching Polymorphic Encryption for Privacy Preserving Machine Learning
Rawan Hagag, Hassan Nassar, Jörg Henkel, Mohamed A. Abd El Ghany · 2025
This paper presents a hardware-accelerated polymorphic encryption framework for privacy-preserving machine learning. The approach employs a mode-switching polymorphic encryption scheme, enabling secure training and inference on data without decryption. We prototype a hardware-accelerated decision tree classifier that efficiently traverses encrypted data stored in FPGA BRAM, ensuring low-latency inference. Experimental results demonstrate the impact of encryption on classification accuracy and computational efficiency, providing a scalable solution for secure outsourced machine learning training.