PNST 945-2024 Artificial intelligence. Technical Structure for Deep Neural Network Model Separation

History

2024: Standard Approval

In early November 2024 Russia , the preliminary national standard (PNST) 945-2024 was adopted. " Artificial intelligence Technical structure for the separation and joint execution of the deep neural network model. "

The document was developed by the Scientific and Educational Center for Competencies in the Field of Digital Economy of the Federal State Budgetary Educational Institution of Higher Education "Moscow State University named after M.V. Lomonosov" (FSBEI HE Moscow State University named after M.V. Lomonosov) and Limited Liability Company "Institute for the Development of the Information Society" (LLC "IRIO").

𝓲

Freepik

It is noted that the inference process for a deep neural network model usually requires large amounts of computing resources and memory. In this regard, the independent execution of the model by endpoints is difficult. An effective way to solve the problem is to implement joint execution of a deep neural network by several devices by dividing the model: this approach allows you to reduce execution delay and simultaneously improve resource use.

The new standard aims to define the technical structure of separation and joint execution of a deep neural network model, including creating a delay prediction model, developing a separation strategy, optimizing resource allocation, and joint execution to meet model inference requirements and ensure efficient resource utilization between endpoints and peripherals. The document defines several typical scenarios for the separation and joint execution of a deep neural network model: this is a smart home, industrial production, intelligent transport systems, etc.^[1]

Notes