Lobachevsky University wins major research projects contest
In the contest of research projects announced by the Russian Ministry of Science and Higher Education, only 41 out of 376 applications were supported after a competitive selection process. This number includes the project from Lobachevsky University «Reliable and logically transparent artificial intelligence: technology, verification and application in socially significant and infectious diseases». The project is aimed at creating a new generation of artificial intelligence (AI) and machine learning systems capable of detecting and quickly correcting errors. These correction procedures will involve the implementation of logically understandable solutions.
«First and foremost, the use of these systems is needed in sensitive areas, such as biomedical applications, where people's lives depend on the decisions made,» says Mikhail Ivanchenko, project coordinator and UNN Vice-Rector for Research.
The project's principal investigator is Professor Alexander Gorban, a leading scientist and head of the megagrant laboratory at Lobachevsky University that focuses its research on artificial intelligence and machine learning. The main result of the project should be the development of new methods and technologies to overcome two major barriers to machine learning systems and artificial intelligence: the problem of errors and the problem of explicit explanation of solutions. To date, these problems have not been\ satisfactorily solved and require new research.
«These problems are closely connected: without having a possibility of logical rendering, the errors of artificial intelligence will remain unexplained. Additional learning of the system within the limits of existing methods can damage available skills and, on the other hand, can demand huge resources, which is impractical in serious tasks. For example, the widely known multimillion-dollar IBM AI system "Watson" has failed in the market of personalized medicine due to systematic errors in diagnosis and recommendation of cancer treatment. The sources of such errors could not be found and eliminated,» explains Alexander Gorban.
The new technology of reliable and understandable neural-network based AI will be implemented for a wide list of strategically important tasks with real-life complexity. In each of them, the requirement of reliability and logical explainability is critical, and in each of them the achievement of the following practically significant results is expected: the analysis of large biomedical data and identification of ultra early disease predictors, analysis of climatic data and prediction of extreme events, engineering of new materials, quantum and optical technologies, development of neural networks that implement information-computing (intellectual) functions of the brain.