IBM the release of the source code of the toolkit (IBM Fully Homomorphic Encryption) with the implementation of the for processing data in an encrypted form. FHE allows the creation of services for confidential computing, where data is processed in an encrypted state and does not appear in plaintext at any stage. The output is also generated in an encrypted form. The code is written in C++ and is licensed under MIT. In addition to the version for Linux, similar toolkits are being developed concurrently for and , written in Objective-C. A version for .
FHE supports homomorphic operations, allowing the addition and multiplication of encrypted data (i.e., enabling any arbitrary computations) while producing an output that is encrypted and analogous to the encryption of the result of adding or multiplying the original data. Homomorphic encryption can be seen as the next evolution of end-to-end encryption — in addition to securing data transmission, it allows for data processing without decryption.
Practically, the framework could be useful for organizing confidential cloud computing, in electronic voting systems, in anonymized routing protocols, for encrypted request processing in DBMS, and for confidential training of machine learning systems. An example of FHE application is the organization of patient information analysis in medical institutions by insurance companies without the insurance company gaining access to data that could identify specific patients. Additionally, the development of machine learning systems to detect fraudulent credit card transactions based on processing encrypted anonymous financial transactions.
The toolkit includes a library with the implementation of several schemes of homomorphic encryption, an integrated development environment (work is carried out via the browser), and a set of examples. To simplify deployment, ready-made Docker images based on CentOS, Fedora, and Ubuntu have been prepared. Instructions for building the toolkit from source code and installing it on a local system are also available.
The project has been developing since 2009, but only now has it achieved acceptable performance metrics that allow for practical use. It is noted that FHE makes homomorphic computations accessible to everyone; with FHE, regular corporate programmers can perform tasks in minutes that used to take hours or days when involving experts with advanced degrees.
Among other developments in the field of confidential computing, we can note with the implementation of methods of , allowing for statistically significant operations on datasets without the possibility of identifying individual records within them. The project is jointly developed by researchers from Microsoft and Harvard University. The implementation is written in Rust and Python and is licensed under the MIT License.
Analysis using differential privacy methods enables organizations to produce analytical samples from statistical databases, preventing the isolation of parameters related to specific individuals from the overall information. For instance, to identify differences in patient care, researchers can provide data that allows comparison of the average length of stay for patients in hospitals while maintaining patient confidentiality and preventing the identification of personal details.
To protect identifiable personal or confidential information, two mechanisms are applied: 1. Adding a small amount of statistical "noise" to each result, which does not affect the accuracy of the extracted data but masks the contribution of individual data elements.
2. The use of a privacy budget that limits the amount of data retrieved for each request and prevents additional requests that could compromise confidentiality.
Source: opennet.ru
