The article discusses the stages of entity allocation for constructing an ANN description language that simplifies the exchange of ANN between people. The developed language reduces the task of migrating the ANN between different hardware and software technologies to the task of converting a text description to the required platform, which is much easier to develop intelligent models. As a result of minimizing the set of entities, a language consisting of a dozen and a half tags was obtained that allow describing the most popular ANN models used for classification and pattern recognition tasks.

Keywords: artificial neural networks, neuron, synapse, multilayer perceptron, INSML, forward calculation, mathematical modeling