October 2022 Summaries

Zama has released major updates to four of its libraries, enhancing its suite of products aimed at making fully homomorphic encryption (FHE) more accessible, easy, and fast. Concrete Core v1.0 offers a platform for integrating FHE primitives on any hardware, focusing on improving performance through hardware acceleration. Concrete v0.2 builds on Concrete-core v1.0 by deprecating certain crates, optimizing Fourier transforms, parallelizing key generation, and supporting Apple Silicon architectures. Meanwhile, Concrete Numpy v0.8 has been rewritten to reduce technical debts and now supports Python 3.7 and 3.10, large bit width support, and tensor assignments. Lastly, Concrete ML v0.4 introduces support for quantization aware training neural networks, decision tree-ensemble regressors, and additional linear regression models, with first-class support for Brevitas, a quantization aware training framework.

Zama has announced significant updates in its Concrete-Numpy library from version 0.6 to 0.8, focusing on improving the functionality and efficiency of encrypted computations. The library was restructured in version 0.6 to reduce technical debt and introduced breaking changes like renaming components and incorporating Virtual Circuits for prototyping. Key enhancements include the ability to split circuits into Client and Server components for deployment, configure per Table Lookup error probabilities, and perform univariate functions more easily. Version 0.7 introduced support for signed inputs, expanded capabilities for creating tensors filled with zeros and ones, and improved error messaging. The latest update, version 0.8, includes support for Python 3.7 and 3.10, enhanced bit-width support, native subtraction between encrypted values, improved invalid value detection during inputset evaluation, and extended NumPy function support, such as grouped 2D convolutions and assignment to tensors. These updates aim to streamline encrypted data processing and broaden the library's applicability across various platforms and services.

Zama has announced a new version of Concrete, following the concrete-core V1.0.0 release, which introduces an enhanced high-level interface for developers through three main components: concrete-boolean, concrete-shortint, and concrete-integer. The update includes several significant features such as the deprecation of the concrete-core-experimental crate, parallelized key generation that accelerates performance by utilizing available CPU cores, and native support for Apple Silicon using Rust's nightly compiler. Notably, concrete-boolean now supports GPU-accelerated homomorphic operations via concrete-cuda, and concrete-integer introduces the WoPBS cryptographic approach for evaluating functions over large integers without padding, supporting univariate functions on 16-bit integers and bivariate functions on smaller integers. The API updates include dedicated ciphertext types like RadixCiphertext and CrtCiphertext, ensuring operational compatibility and offering automatic parallelization for enhanced performance. These advancements allow for versatile and efficient computations over integers in the Concrete frontend while maintaining consistent syntax and improving performance without requiring changes to existing programs.

Zama has released a new version of Concrete-ML, featuring client/server APIs for deployment, new machine-learning models, improved processing speed, and support for Quantization Aware Training (QAT). The client/server APIs enable key generation, data encryption, model execution on untrusted servers, and result decryption on clients, facilitating production deployment of models. The release expands available machine-learning models, adding regressors like Lasso, Ridge, ElasticNet, and tree-based models like DecisionTree and RandomForest. An important advancement is the introduction of QAT, which enhances model accuracy by optimizing weights under low bit-width constraints, making models more effective compared to previous versions. This update includes integration of QAT into built-in neural network models and the ability for users to import custom quantized models, with Brevitas being used for quantization on datasets like MNIST. Future efforts will focus on tackling complex tasks using QAT and extended precision in computation.

Zama has released Concrete-core V1.0.0, an open-source platform aimed at advancing Fully Homomorphic Encryption (FHE) by improving performance and ease of use. This version focuses on integrating hardware acceleration into the Concrete Framework, utilizing a Rust library with a plugin system for hardware contributions, a comprehensive testing and benchmarking framework, and a tool for automatic API generation to other languages. It includes CPU implementations of TFHE operations, Cuda acceleration for key processes, and a flexible architecture that allows for the easy addition of new hardware backends. Concrete-core V1.0.0 aspires to bridge the gap between hardware developers, cryptographers, and software developers to facilitate FHE's broader adoption. However, the platform remains research-intensive, with ongoing developments anticipated to enhance hardware integration further, notably through the Concrete Compiler, which aims to make FHE more accessible to users of Concrete-Numpy and Concrete-ML.

Post-Quantum Cryptography (PQC) has garnered significant attention due to concerns about quantum computers potentially breaking existing cryptographic systems like RSA and ECC, which are crucial for modern security. To address this, the National Institute for Standards and Technology (NIST) has embarked on creating new algorithms resistant to quantum decryption, selecting Kyber for encryption and Dilithium, Falcon, and SPHINCS+ for signatures as part of their formal standardization process in July 2022. These selections predominantly involve lattice-based cryptography, which is believed to withstand quantum computing challenges due to its reliance on complex mathematical problems unsolvable by quantum algorithms like Shor's. Lattice problems involve finding short vectors to generate a lattice, which becomes increasingly difficult with higher dimensions, providing a robust foundation for cryptographic standards. This advancement also benefits homomorphic encryption schemes, such as the TFHE scheme by Zama, which utilize lattice-based cryptography and the Lattice Estimator tool, showcasing a collaborative effort to ensure secure cryptographic solutions in a post-quantum world.