Reliability Modeling And Evaluation Method Of AI Component System Based On Composite Semantics

With the continuous development and progress of artificial intelligence technology,in the process of software development,more and more machine learning components are integrated into the software system,use intelligent and data-based methods to complete specific tasks.The software system of the learning component is also known as the AI system.On the one hand,because the machine learning component has the problem that the probability of getting the correct result is a statistical value and the process of obtaining the result is difficult to explain,the method of traditional software engineering modeling and evaluation can no longer fully meet the needs of the development and evaluation of intelligent software systems.The traditional system reliability evaluation method relies on the reliability of each component itself and the influence of the cooperation between components on the reliability.Research on the impact of collaboration on reliability is still in its infancy.To solve the above problems,this paper proposes a reliability modeling and evaluation method for AI systems,which is used to model and evaluate the reliability of AI systems.First,this paper proposes a formalized probabilistic semantic representation of AI component reliability,which is used to model the reliability of machine learning components based on probabilistic approximate correctness.Then a description language of component combination operator is proposed,which is used to describe the interaction pattern between components,and calculate the reliability of the whole system according to the semantic model of the interaction pattern.Finally,combined with the open-source automatic driving system,the framework proposed in this paper is used to conduct a case study and explanation of the traffic light recognition module of automatic driving.The main work and contributions of this paper are as follows:1.In order to model and represent the reliability of AI components,this paper proposes an AI component reliability specification representation method based on probabilistic semantics,which is used to describe the reliability of components in AI systems.For modeling the reliability of machine learning components based on probabilistic approximation to correctness.2.In order to formally describe the interaction and collaboration behavior between machine learning components and non machine learning components,the text proposes a description language of collaboration mode between components,which describes the combined structure of the system in the way of graphical language,and provides its formal semantics and reliability calculation methods.Based on this framework,the reliability of the entire AI system can be calculated given the combination structure between the components of the system and the reliability of each component(AI component and non AI component)has been evaluated.3.In order to verify the practicability and easy scalability of the proposed method,based on the above work,the method proposed in this paper is used to actually model the traffic light perception module in the apollo autonomous driving system,and the neural network model is actually trained to construct the traffic light recognition process to Obtain the reliability value of each component,and calculate the reliability of the modeling system according to the combined structure of the components.