Over the past few years, a tremendous growth of machine learning was brought about by a significant increase in adoption and implementation of cloud-based services. As a result, various solutions have been proposed in which the machine learning models run on a remote cloud provider and not locally on a user's machine. However, when such a model is deployed on an untrusted cloud provider, it is of vital importance that the users' privacy is preserved. To this end, we propose Learning in the Dark -- a hybrid machine learning model in which the training phase occurs in plaintext data, but the classification of the users' inputs is performed directly on homomorphically encrypted ciphertexts. To make our construction compatible with homomorphic encryption, we approximate the ReLU and Sigmoid activation functions using low-degree Chebyshev polynomials. This allowed us to build Learning in the Dark -- a privacy-preserving machine learning model that can classify encrypted images with high accuracy. Learning in the Dark preserves users' privacy since it is capable of performing high accuracy predictions by performing computations directly on encrypted data. In addition to that, the output of Learning in the Dark is generated in a blind and therefore privacy-preserving way by utilizing the properties of homomorphic encryption.

翻译：在过去几年中，云服务的大规模采用与部署推动了机器学习的飞速发展。由此催生出的多种解决方案中，机器学习模型运行在远程云提供商而非用户本地设备上。然而，当此类模型部署于不可信的云平台时，保护用户隐私至关重要。为此，我们提出"暗处学习"——一种混合机器学习模型：训练阶段对明文数据进行处理，而用户输入的分类则直接在同态加密的密文上执行。为使我们的架构兼容同态加密，我们使用低次切比雪夫多项式逼近ReLU和Sigmoid激活函数。这使我们得以构建"暗处学习"——一个能够对加密图像进行高精度分类的隐私保护机器学习模型。该模型可直接对加密数据进行计算实现高精度预测，从而保护用户隐私。此外，通过利用同态加密的特性，"暗处学习"的输出以盲处理方式生成，进而实现隐私保护。