FebriantiW/Homomorphic-Encryption-and-Federated-Learning-based-Privacy-Preserving-CNN-Training-

This project is a classic academic reference implementation for a research paper. While it addresses a critical intersection (FL + HE for medical AI), it lacks the engineering maturity required for production environments. With only 23 stars over nearly four years and zero recent velocity, it has failed to gain traction as a library or tool. The defensibility is near zero because the core value lies in the mathematical approach described in the paper, which can be (and has been) implemented more efficiently in industrial-grade frameworks. In the current market, this project is effectively displaced by mature ecosystems like OpenMined's PySyft, Flower, or NVIDIA FLARE, which provide robust, optimized, and maintained versions of these exact privacy primitives. Furthermore, the use of Homomorphic Encryption for CNN training—while theoretically sound—suffers from massive computational overhead that modern projects mitigate through more advanced techniques like Secure Multi-Party Computation (SMPC) or hardware-based TEEs (Trusted Execution Environments). Frontier labs and cloud providers (Azure Confidential Computing, GCP) are rapidly absorbing these capabilities into platform-level offerings, making stand-alone, unmaintained academic repos like this one obsolete for all but historical reference or specific algorithmic inspiration.