October 2023 Summaries

Zama's fhEVM introduces confidential smart contracts to blockchain technology by employing fully homomorphic encryption (FHE), which allows transaction data and on-chain states to remain encrypted throughout processing, thereby unlocking new applications in areas like decentralized finance (DeFi) and gaming. Traditional blockchains struggle with confidentiality due to the public nature of on-chain data, but Zama's innovative protocol addresses this challenge. A video tutorial by blockchain engineer Clément Danjou provides guidance on how to write confidential smart contracts using fhEVM, and the initiative is supported by a GitHub repository, comprehensive documentation, community support channels, and a bounty program aimed at advancing the FHE space.

The text explores the trade-offs between cloud and on-premise deployment of large language models (LLMs), highlighting the challenges of data privacy, model IP protection, and compliance with licensing agreements. It introduces a hybrid approach using Concrete ML's Fully Homomorphic Encryption (FHE), which allows sensitive data processing to be offloaded to the cloud while preserving user privacy and maintaining the on-premise performance. The hybrid model leverages the HybridFHEModel class to convert PyTorch models into hybrid ones, enabling some layers to run on encrypted data remotely. This approach aims to balance user privacy, model security, and operational efficiency, illustrated by benchmarking the phi-1.5 model, showing a generation latency of 2.5 seconds per token compared to 50ms for fully on-premise setups. Future improvements in Concrete ML aim to reduce latency and data transfer sizes through innovations like ciphertext seeding and GPU utilization for server-side computation.

Zama's fhEVM introduces confidential smart contracts to blockchain technology, addressing the challenge of maintaining privacy in decentralized applications where on-chain data is inherently public. By utilizing fully homomorphic encryption (FHE), Zama's fhEVM ensures that transaction data and the on-chain state remain encrypted throughout processing, thus unlocking new use cases in sectors like DeFi and gaming. This innovative protocol is demonstrated by team member Louis Tremblay-Thibault, who provides guidance on using the CMUX Operator on Zama's platform. The initiative also encourages community engagement through their GitHub repository, documentation, support channels, and incentive programs like the Zama Bounty Program & Grant Program, aimed at advancing FHE technology.

Concrete is an open-source fully homomorphic encryption (FHE) compiler designed to ease the implementation of FHE by allowing computations on encrypted data without decryption. It integrates a TFHE Compiler based on LLVM, providing a user-friendly way for developers to write FHE programs. A tutorial by Zama team member Umut Sahin demonstrates the use of dynamic table look-ups with Concrete, and the project encourages community engagement through its GitHub repository, documentation, and support channels. Additionally, Zama promotes the advancement of the FHE field through initiatives like the Bounty Program and Grant Program.

Zama has released new versions of its homomorphic encryption products: TFHE-rs v0.4.0, Concrete v2.4.0, Concrete ML v1.2.0, and fhEVM v0.2.0, each introducing significant advancements. TFHE-rs v0.4.0 now supports signed integers and ternary homomorphic conditional instructions, alongside performance improvements. Concrete v2.4.0 enhances FHE efficiency with multi-parameter optimization, improved statistics reporting, and dynamic table look-ups, resulting in computations up to five times faster. Concrete ML v1.2.0 offers hybrid deployment for model IP protection, K-nearest neighbor classification for encrypted data, and 10x faster neural networks due to optimized input division, with updates expedited via a public repository. fhEVM v0.2.0 expands its capabilities with a comprehensive set of operators for greater functionality and versatility. These updates underscore Zama's commitment to making homomorphic encryption more accessible, efficient, and rapid.

Concrete ML version 1.2 introduces several advancements, including hybrid deployment with Fully Homomorphic Encryption (FHE) and support for K-nearest neighbor (KNN) classification, aimed at enhancing model security and deployment flexibility. The hybrid deployment allows models, including Large Language Models (LLMs) and Convolutional Neural Networks (CNNs), to be partially deployed on-premise and in the cloud, protecting both confidential data and model intellectual property. This version also incorporates optimizations in neural networks, such as right bit-shift and quantization-aware training, significantly improving inference speed while maintaining accuracy. The KNN classification is performed on encrypted data, ensuring data privacy through encrypted distance calculations. Additionally, the development pipeline has transitioned to a public repository to expedite bug fixes, enabling developers to access the latest code improvements.

The latest version of TFHE-rs brings notable enhancements, including support for signed integers, conditional instructions, and significant performance improvements. The update introduces a new type, FheInt, which supports signed integers similarly to the existing FheUint type for unsigned integers, extending operations like addition, subtraction, and multiplication to signed numbers with comparable timings. Conditional instructions now allow simple homomorphic computations using encrypted boolean values in If-Then-Else constructs, enabling dynamic selection between expressions based on encrypted conditions. Performance improvements are evident, with an overall latency reduction of 8% across integer operations, and specific optimizations like a 30-50% speed increase in multiplication between clear values and ciphertexts, and nearly double the speed for division operations. The update also includes enhancements to low-level primitives, particularly programmable bootstrapping, contributing to the improved efficiency of the framework.

Zama's recent updates to its fully homomorphic encryption Ethereum Virtual Machine (fhEVM) have introduced a range of new operators and supported encrypted data types, enhancing functionality and versatility. The latest version, v0.2.0, now supports encrypted boolean and unsigned integers, with plans to include signed types and 64-bit integers soon. Operators for arithmetic, bitwise operations, bit shifting, and comparisons are now available, with special operations such as the CMUX operator and random encrypted integer generation also introduced. These updates enable more secure and efficient on-chain computations while maintaining data privacy. The platform advises caution when using certain functions in production due to potential security risks and emphasizes optimizing performance with the optReq method, which reduces transaction flow interruptions by accumulating conditions for batch decryption. Zama's specialized Remix instance and Hardhat template facilitate the development and deployment of applications using fhEVM, while community engagement is encouraged through GitHub, documentation, community channels, and a bounty program aimed at innovative uses of programmable privacy.

Since the release of Concrete v2.0.0 in July 2023, four minor updates have been introduced, enhancing performance, adding new features, and fixing bugs. A major improvement is the multi-parameter optimization, which uses several key-switching and bootstrapping keys to improve the efficiency and latency of Fully Homomorphic Encryption (FHE) evaluations. This optimization, active by default from version 2.1.0, has improved the speed of various operators. The updates also include the provision of detailed statistics for developers, offering insights into cryptographic operations to facilitate better program optimization. Additionally, a new feature enables dynamic table lookups, allowing computations with tables determined at execution time. These enhancements, alongside other minor improvements, have significantly boosted the performance of Concrete, as evidenced by benchmarks like CIFAR, which now run markedly faster.

The blog post explores the use of homomorphic encryption to create confidential Decentralized Autonomous Organizations (DAOs) using Zama's fhEVM, focusing on transforming a traditional DAO into a confidential one. A DAO, typically governed by smart contracts such as the Governor and Token contracts, allows token holders to propose and vote on organizational decisions, but in a traditional setup, voting records and token holdings are public. A confidential DAO, however, maintains the secrecy of token quantities and individual votes by employing encrypted data types like euint32 and ebool. The post details the process of converting the COMP token contract into a confidential token contract and making necessary adjustments to the Governor contract to ensure vote confidentiality using encrypted integers. It highlights the ease of transitioning an unencrypted DAO contract to an encrypted one with fhEVM, which facilitates writing smart contracts with encrypted data and performing homomorphic computations. The article provides insights into managing historical data for token holdings, vote secrecy, and proposal thresholds in confidential DAOs, encouraging readers to explore further through documentation, community support, and a bounty program.

Concrete, Zama's open-source Fully Homomorphic Encryption (FHE) compiler, aims to make FHE accessible beyond cryptography experts by providing a modular framework that can be extended and customized. The compiler, which uses MLIR for efficiency, translates plaintext programs into FHE programs through Concrete Compiler and supports user-friendly development via Concrete Python, a frontend that allows Python-written programs to be converted into MLIR-based intermediate representations. This setup is designed to facilitate the integration of FHE into applications without requiring developers to write complex code, with the potential for additional language frontends to emerge from the open-source community. Concrete also introduces Concrete ML, which adds privacy features to machine learning frameworks like scikit-learn and Keras. The compiler employs a high-level approach to encoding, optimizing the translation of arithmetic circuits into TFHE equivalents, and leverages parallelism for hardware efficiency. The ongoing standardization of MLIR-based FHE dialects through the HEIR project and community involvement, such as the Zama Bounty Program, are pivotal to its development and adoption, with the project already gaining significant community support and GitHub recognition.

Concrete is Zama's Fully Homomorphic Encryption (FHE) Compiler, designed to convert Python programs into their FHE equivalents, enabling developers to create applications that handle encrypted inputs and outputs. In a video tutorial, Zama team member Ayoub Benaissa provides an in-depth exploration of Zama's TFHE compiler, which utilizes MLIR. The text also encourages engagement with the Concrete project by starring its GitHub repository, reviewing its documentation, and participating in the community channels. Additionally, it highlights the opportunity to learn more about FHE and contribute to advancements in the field through the Zama Bounty Program, which offers monetary incentives.