Research On Refinement Learning For Resource Network With Application

The real world of human life is already the Internet of resources fusion in many fields.Resources in the Internet environment is a very broad concept as it involves many areas. Either objective or subjective structures,anything that can be identified is a kind of resource at a particular domain.Resource network is a structure of tree,lattice or graph that is expressed through concept,relationship,and rule.Resource network is a knowledge model that is established from thinking of human brain for resource in the Internet environment.Resource modeling method in this dissertation is achieved through the refinement learning for resource network.Refinement learning for resource network includs three sub-task procedures of the concept induction,rule induction and program evolution,and the procedures is verified through the application case of formal concept,the first-order concept induction and evolution.The application cases show that refinement Learning is a search process for problem solving.Problem space and solution space are described by logical rule grammars,and solving process using derivation tree search strategies.In this dissertation,refinement learning methods for the resource network will provide a new machine learning method for resource modeling in various fields,such as Grid,Cloud computing and Internet of Things.Application cases and experimental tests show that this research has theoretical and practical value.In this dissertation,refinement learning with applications for resource network has been systematically researched,and make the following contributions:It is proposed or established that Logic Rules Grammar,Granularity Formal Concept Analysis,Deduction Inductive Logic Programming,Logic Genetic Programming,and Refinement Logic Genetic Programming.Logic Rules Grammar is a expansion of traditional Definite Clause Grammar. Granularity Formal Concept Analysis is a expansion of traditional Formal Concept Analysis.Deduction Inductive Logic Programming is a expansion of traditional Inductive Logic Programming.Logic Genetic Programming is a expansion of traditional Genetic Programming.Refinement Logic Genetic Programming is a expansion of Logic Genetic Programming.By Logic Rule Grammars,it has been integrated into refinement learning approach that three learning paradigm with symbols,connections and evolutionary characteristics about the concept induction,rule induction and program evolution for the resource network,and applied to refinement learning area about formal concept,first-order concept. It is to provide a new machine learning methods with a fusion thinking of symbols,connection and evolutionary for resource modeling to network computing.The concrete contents of this dissertation about five innovation points are given as below:1. From thinking of the formal language,the method is proposed resource structure description to view grammar that is a logic rule grammar(LRG) for resource network.2. From thinking of the formal concept,the method is proposed concept refinement to view granularity that is concept induction(GFCA) for resource network.3. From thinking of the symbol logic,the method is proposed concept refinement to view deduction that is rule induction(DILP) for resource network.4. From the biological evolution of thinking,the approach is proposed concept refinement to view evolution that is program evolution(LGP) for resource network.5. Using logical rule grammars,the three concept refinements are integrated into a learning approach that is refinement learning approach(RLGP) for resource network.