May 2023 Summaries

Concrete ML is a set of privacy-preserving machine learning tools designed to simplify the use of Fully Homomorphic Encryption (FHE) for developers, enabling the conversion of machine learning models into their homomorphic equivalents. The release of Concrete ML v1.0.0 brought enhancements like improved performance and better model development assistance. A focus of the tool is deploying models, such as a breast cancer classification model, to AWS EC2 using FastAPI servers, facilitated by utility scripts that ease the deployment process via a command-line interface leveraging Boto3. The deployment involves creating an AWS EC2 instance, transferring files, installing dependencies, and running the server, with logs providing the URL for client access. Users can develop client applications to interface with the server, and the documentation provides guidance on using Client/Server APIs and addressing deployment issues, including potential future support for AWS ECR and ECS. Developers are encouraged to explore the documentation, contribute to the GitHub repository, and participate in the Zama Bounty Program to advance the field of FHE.

During its second season, the Zama Bounty Program awarded €29,000 to developers for solving various challenges related to Fully Homomorphic Encryption (FHE) and Zama's libraries, with successful projects including tutorials for SHA256, dark market applications, and homomorphic regex engines, among others. As the program transitions into its third season, it introduces a streamlined approach by focusing on one bounty per Zama library, aiming to boost participation and diversify submissions with increased rewards for the top three entries, with cash prizes ranging from €1,500 to €10,000. Participants are encouraged to propose their own bounties, and the season will continue until August 27, 2023, with submissions being judged on code quality and performance speed, tested on Amazon EC2 M6i instances.

Zama has released the alpha code and white paper for the fhEVM, showcasing the potential of Fully Homomorphic Encryption (FHE) in enabling private smart contracts on public blockchains. FHE allows encrypted data to be processed without decryption, providing privacy by keeping transaction data and contract states encrypted. Recent advancements have made FHE practical, paving the way for applications such as trustless bridges, blind auctions, and confidential voting. Despite challenges like noise growth and access control, techniques like bootstrapping and schemes like TFHE are addressing these issues, offering improved privacy and computational capabilities. With ongoing developments, including FHE-rollups and hardware accelerators expected by 2025, FHE smart contracts are set to achieve significantly higher throughput, making private blockchain interactions feasible.

Zama's Concrete is a fully homomorphic encryption (FHE) compiler designed to simplify the complexities of FHE for developers by handling noise management, crypto parameter selection, and operation ordering. Utilizing MLIR (Multi-Level Intermediate Representation), the compiler expects input in MLIR format and supports various APIs, including Python, C++, and C, along with a CLI tool for debugging. The Python API offers an accessible setup for compiling and running FHE programs, such as performing encrypted additions and matrix multiplications. The compilation process involves generating keys, encrypting inputs, executing the compiled function, and decrypting results, with crypto parameters automatically determined by the compiler. Concrete Python serves as a frontend layer, translating user input into MLIR, though the open-source nature of Concrete encourages the development of custom frontends. The Concrete Compiler facilitates the creation of encrypted computations without requiring extensive cryptographic knowledge, supporting further exploration and innovation in the FHE space.