Abstract
The article proposes a methodology for creating a system for evaluating the quality of artificial intelligence (AI) solutions, in particular, evaluating the trustworthiness of solutions, which determines the degree of confidence in the solutions obtained. The methodology is based on the regularizing Bayesian approach and implemented as a system of metrological support of AI solutions. Within the framework of this methodology and technologies of metrological support, complexes of metrological characteristics are proposed that determine accuracy, reliability (level of errors of the 1st and 2nd kind), reliability, risk, entropy, amount of information for each stage of the algorithms, which ensures traceability and transparency of the solutions obtained. Practical examples of determining metrological complexes for solving applied artificial intelligence systems are given.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have