The seminar was organized by Professors A.N. Karapetyants and V.V. Kravchenko. A team of scientists from MIPT presented their developments. An analysis of approaches to the problem of detecting machine-generated text fragments based on the use of transformer models to improve recognition quality was presented.

The speakers focused on the development of improved methods aimed at identifying generation features at various levels of the text structure: from words and sentences to logical blocks within the document. Multitasking learning approaches that provide regularization of feature representations of text fragments and increase the generalizing ability of models were covered, and a comprehensive analysis of data sets used in this field for their quality and informativeness was presented. After the presentation, a discussion and exchange of opinions was held with the participation of potential customers.