Concrete ML by Zama is a privacy-preserving machine learning framework built on top of Concrete, with bindings to traditional machine learning frameworks. It supports the conversion of pre-trained linear models through from_sklearn and pre-trained neural networks using compile_torch_model. Concrete ML enables the training and inference on encrypted data without decryption, maintaining confidentiality throughout the AI processing lifecycle. It introduces a new DataFrame API for working with encrypted stored data, enhancing private inference and confidential collaboration. Concrete ML supports the training of logistic regression models directly on encrypted data, ensuring data security while facilitating collaboration. The framework also includes features for encrypted training and faster decision trees, allowing for the training of models on encrypted data without needing to decrypt it first. Concrete ML improves latency on large neural networks and supports pre-trained tree-based models, making it suitable for complex machine learning tasks. The framework is designed to make homomorphic encryption accessible, easy, and fast for developers, enabling them to build models that run on encrypted data without requiring extensive cryptographic knowledge.
