The Analysis Of Knowledge Representation In Terms Of Non-classical Logic

Machine intelligence has always been the aim of artificial intelligence,and machine learning as an important branch of artificial intelligence is a vital theme of the research of machine intelligence.But machine learning has a dilemma,that is,the machine can't represent all human reasoning and real world facts.That is because of the complexity and diversity of the real world,many knowledge is uncertain and fuzzy.Therefore,how to enhance the representational ability of the machine for the representation and processing of the different types of knowledge(i.e.,“the problem of machine learnability”)has been an urgent problem for computer research.Non-classical logic,as a new representation form,which retains the formal characteristic of classical logic and gives up the necessity reasoning of classical logic,is the most potential research method for the current machine learning.Here,I want to emphasize that the machine is a computer system which is formed by software and hardware,rather than a simple mechanical device.Based on the perspective of non-classical logic,the dissertation explores the structure characteristic of non-classical logic,analyzes how to represent the various facts and knowledge on real world from the formal structure and reasoning technique of non-classical logic.The dissertation mainly is divided into three parts,which chapter 1 is the first part,chapter 2,3,4,5 are the second part,chapter 6 is the third part.The first part mainly explores the relation and difference between the concepts of machine learnability,investigates the characteristics of the current research of machine learnability.On this basis,the analysis of the characteristics of non-classical logic indicates that non-classical logic is the most potential method for the research of machine learnability,and this method has strong formal reasoning ability and the uncertainty of reasoning which guarantees the validity of reasoning and enriches the reasoning results.The second part analyzes the four types of the knowledge which have different characteristics in detail in terms of different logic systems,provides a potential research way for the representation of uncertain knowledge.The second chapter compares possible world semantics and situation semantics,clarifies that possible world can been interpreted as situation,and points out that modal logic has similar characteristics with situation semantics,so modal logic can be used for the representation of situation knowledge.The third chapter reveals that context is the key for the solution of fuzzy problem,therefore,it suggests solving fuzzy problem by the clarification of context,and analyzes concretely fuzzy problem based on two different kinds of context logic.The fourth chapter analyzes meaningful contradiction in detail,expounds that paraconsistent logic can deal with meaningful contradiction,and elucidates the analytic process of contradiction by four-value structure and the structure of annotated lattice.Then,the fifth chapter,as the analysis of incomplete empirical knowledge,reflects the credibility and validity of empirical knowledge by computing the probability of favourable evidences basedon the interpretation of the credibility of probability theory,further analyzes the revision process of empirical knowledge in detail.The third part is the philosophical reflection on the representation of uncertain knowledge.From the perspective of logic,it reveals the change of knowledge representation problem in research content and research method,indicates that non-classical logic is one method which is the combination of syntax analysis and semantic analysis and the integration of context and computation,and the method also is trying to achieve the unity of scientific rationality and humanist rationality.And from the perspective of epistemology,knowledge is diversified because of context and evidence relevance,and points out that knowledge can be mistaken.Conclusion summarizes that the research of machine learning and artificial intelligence will tend to be computable from the representation of uncertain knowledge,the demand of the analysis of big data and the revolution of formal method,and points out that computationalism will be the future trend of development,further briefly analyzes the development of computationalism from physical architecture,research patterns,research objects and research methods.On this basis,conclusion illustrates that commonsense reasoning,natural language,biologically motivated models and the sample data of machine learning will be the problems which should be researched in the future.In a word,the dissertation firstly discusses the concept,characteristics and research methods of machine learnability.On this basis,the dissertation further investigates the representation of various,fuzzy and uncertain knowledge by non-classical logic,and elucidates the adaptation and rationality of non-classical logic in the solution of the problem of knowledge representation.The purpose of dissertation is to suggest that non-classical logic is rational for different types of knowledge representation,non-classical logic is an important way for the research of machine learnability,is also a novel ways for the research of machine intelligence,which has important scientific value and philosophical significance.