February 2023 Summaries

Zama has developed a new Hugging Face space utilizing Concrete-Numpy and Concrete-ML to apply homomorphic filters over images, ensuring data remains encrypted during transit and processing. Concrete-Numpy enables computations on encrypted data without decryption, while Concrete-ML allows the application of machine learning models in a Fully Homomorphic Encryption (FHE) setting without requiring cryptographic expertise. Users can create image processing filters, such as sharpening or noise reduction, using Torch models and compile them into FHE circuits. These filters can be deployed via a Client-Server interface, maintaining privacy by encrypting images before processing, with both the client and server components capable of handling encrypted data. A demo on Hugging Face showcases this functionality, where images are encrypted, processed, and decrypted, with the server having no access to the original or processed images. This approach ensures privacy and security, making it suitable for sensitive image processing applications.

During its inaugural run, the Zama Bounty Program awarded over €15,000 to developers, successfully engaging the community in solving both basic and complex issues related to Zama's libraries. Contributors tackled tasks ranging from creating educational tutorials for the machine learning library Concrete-ML to addressing intricate cryptographic challenges like improving TFHE programmable bootstrapping latency. Building on this success, the second season of the program has been launched with increased bounties and an enhanced process, based on participant feedback. The current season offers a diverse array of bounties across five main categories, including Application Engineering and Machine Learning, with over $325,000 in potential earnings. The program encourages community involvement by allowing members to propose new bounties, aiming to foster collaboration and advance the development of open-source tools. Season 2 is set to conclude on April 7th, offering a limited window for participation.