The Use of the Concept of “Typical Process” in Machine Learning and Artificial Intelligence Systems

Abstract

Typical processes are a fundamental invariant of many natural and artificial systems. The purpose of a typical process is to maintain homeostasis in the system. The system is in this case an information control network (ICN) with three generations of objects (the central network object acts as a teacher). Applying the concept of Typical Process and Semantic Anomaly in machine learning and artificial intelligence systems will avoid lots of computations and unpredictable results typical for systems with Deep Learning. While admittedly obvious, the concept of a typical process is difficult to formalize. It is proposed to use the mathematical apparatus of perturbation theory and fast-slow dynamic systems, slow variables are interpreted as semantic. An example of a machine learning problem is considered – the entry of a new object into a system reproducing an aggregated typical process. A possible tool for modeling changes in the typical process under the influence of the external environment is proposed to model using trajectories-ducks. To solve the problems of control of typical processes in automatic systems, the mathematical apparatus of Kurzhansky trajectory tubes and the predictor-corrector scheme, which in this situation has a certain physical meaning, are proposed. Artificial intelligence systems require modeling hierarchies of typical processes, which is a nontrivial mathematical problem.

Keywords

• artificial intelligence
• fast-slow dynamic system
• homeostasis
• information management network
• semantic space
• emantic anomaly
• predictor-corrector scheme
• typical process
• machine learning
• semantic anomaly
• supervised learning
• trajectories-ducks
• tubes of trajectories