其他名称：
25-21-00330
项目负责人：
Losev Alexander
发表日期：
2025
主持机构：
Volgograd State University,
国家：
俄罗斯
开始日期：
2025
结束日期：
2026
简介：
The scientific problem of this Project is related to the need to develop new methods for generating a training data set through the use of artificial intelligence technologies, as well as computer and surrogate modeling.
The methods and models being developed must provide the necessary adequacy and accuracy to satisfy the following requirement: high sensitivity and specificity of classification algorithms and, accordingly, the quality of the proposed diagnostic solution.
The relevance of the study is primarily determined by the most important problem of machine learning associated with the presence of errors, omissions and imbalance of source data. Note that imbalance can be observed both in relation to objects of different classes and in the descriptive characteristics of objects of the same class. What makes this task particularly difficult is the fact that errors can be associated not only with measurement errors, but also with the relevance of the data to the accepted models. This problem has emerged in recent years in almost all areas of application of artificial intelligence methods. On the other hand, it is not known for certain how adequately the model reflects reality. Situations regularly arise when the adopted models contain incorrect data that are far from reality, and this is not due to measurement methods, but to the understanding of the problem. Machine learning methods are capable of extracting knowledge from statistical information. But if there is no knowledge in the data sets presented for training, then machine learning methods are powerless. Another, no less important problem is the small amount of data. With limited sampling, every object is important for machine learning models. And if it is incorrect, the quality of the models drops significantly. At the same time, removing a given object from the sample entails the loss of important information. In this regard, there are corresponding restrictions imposed on methods for creating a training data set.
The research of this Project is devoted to the development of methods that make it possible to highlight the truth contained in microwave radiothermometry data and level out possible errors in a limited data set, as well as supplement the existing set with new knowledge. The proposed approach is based on a synthesis of computer modeling and machine learning methods. At the same time, new methods are being developed for data augmentation and the formation of a training data set in data mining. Namely, it is proposed to develop methods for solving this problem based on constructing a training data set in the problems under consideration using computer and surrogate modeling of the thermal fields of human organs. The methods developed within the framework of this Project are of a general nature and are not limited to thermometric data.
专业领域：
信息技术
语种：
英语
