An open intelligent algorithm adaptive combination method and system based on meta-learning and dynamic pipeline

Abstract

The application discloses an open intelligent algorithm adaptive combination method and system based on meta-learning and a dynamic pipeline. By introducing a type system and an adapter algorithm directed acyclic graph construction mechanism, the application realizes dynamic integration of heterogeneous components in an open algorithm library, and overcomes the limitations of a closed algorithm combination space and a preset process of a traditional AutoML system. By integrating environment perception and runtime performance monitoring into a closed loop decision-making process, and designing a dynamic adjustment mechanism based on bias triggering, the system has the ability to cope with concept drift and resource fluctuations after deployment. By constructing and continuously updating a knowledge base recording the mapping relationship between tasks, environments and performance, the system has the ability to make decisions based on historical experience. This meta-learning mechanism enables the system to reduce repeated search overhead when processing similar tasks, realize continuous evolution of performance, and provide a feasible technical path for solving the continuous learning problem of a general intelligent system.

Images