March 23, 2021, Moscow: Rosenergoatom JSC (part of the Electric Power Division of Rosatom State Corporation), AO CONSIST-OS (a subsidiary of the company) and Tsifrum Private Nuclear Industry Digitalization Institution (Rosatom State Corporation) have completed a pilot project to create an industry-specific system for recording and storing passports of datasets.

A dataset is a collection of data in terms of machine learning problems and their description. A dataset passport contains information about its content, owner and purpose of use, and also allows you to assess its applicability for solving consumer problems, to determine the loading methods and options for subsequent use.

The project was implemented as part of the Rosatom program "End-to-end digital technologies and data management" and is aimed at creating a unified platform for the industry-specific register of datasets, machine learning models, methodologies for solving typical problems in the field of artificial intelligence. The database already contains 12 pilot datasets created by Rosenergoatom and Tsifrum within the framework of the projects that use artificial intelligence and machine learning. The system is undergoing the registration procedure in the Register of Russian software.

“Artificial intelligence and machine learning in particular are new, actively developing technologies in the industry. A large amount of datasets have already accumulated. They are used to train artificial intelligence in various projects. In this regard, Rosenergoatom and the industry as a whole faced the issue of creating a register and realizing the possibility of reusing existing datasets in other projects. This will significantly reduce the time and labor costs for preparing data for new models creation,” - commented Oleg Shalnov, Director of the Department of IT Project Management and Integration (Rosenergoatom JSC).

Each dataset is placed in the registry along with a detailed description of its content, purpose and history of use. This information allows you to assess the potential suitability of a particular dataset for solving other problems and options for its subsequent use. The presence of the register also makes it possible, in the event of failures in the operation of systems with artificial intelligence, to easily find the initial data on which the neural network was trained, to analyze and make the necessary adjustments to the model.

Konstantin Kudashev, head of the Rosenergoatom Digital Technologies Center, emphasized in turn that this system allows to solve the important problem of the safe use of artificial intelligence at industry enterprises. “Safety and efficiency of artificial intelligence systems directly depends on the quality of the data on which the machine learning models are built and trained. All our datasets are verified, tested on real models, and work in industrial systems, which allows us to create more accurate models. Their very storage, located in our reference data center, ensures the safety, security and transparent use of all datasets,” - he said.

The creation of a dataset register is one of the first projects implemented by Tsifrum in development of digital technologies and the culture of using data in the nuclear industry. “The developed product allows us to track the use and usefulness of data, determine responsibility and take into account the contribution of people involved in the development of the field of artificial intelligence, to the development of the nuclear industry. The project demonstrated that by using digital technologies and joining the efforts of participants, data in the industry is a universal asset that can become a “fuel” for both existing and projected business processes,” according to Deputy CEO for End-to-End Digital Technologies and Governance data of ChU "Tsifrum", Anton Zapryagaev.