Mechanizing hypothesis formation is an approach to exploratory data analysis. Its development started in the 1960s inspired by the question “can computers formulate and verify scientific hypotheses?”. The development resulted in a general theory of logic of discovery. It comprises theoretical calculi dealing with theoretical statements as well as observational calculi dealing with observational statements concerning finite results of observation. Both calculi are related through statistical hypotheses tests. A GUHA method is a tool of the logic of discovery. It uses a one-to-one relation between theoretical and observational statements to get all interesting theoretical statements. A GUHA procedure generates all interesting observational statements and verifies them in a given observational data. Output of the procedure consists of all observational statements true in the given data. Several GUHA procedures dealing with association rules, couples of association rules, action rules, histograms, couples of histograms, and patterns based on general contingency tables are involved in the LISp-Miner system developed at the Prague University of Economics and Business. Various results about observational calculi were achieved and applied together with the LISp-Miner system. The book covers a brief overview of logic of discovery. Many examples of applications of the GUHA procedures to solve real problems relevant to data mining and business intelligence are presented. An overview of recent research results relevant to dealing with domain knowledge in data mining and its automation is provided. Firsthand experiences with implementation of the GUHA method in the Python language are presented.

Hypothesis formation is known as one of the branches of Artificial Intelligence, The general question of Artificial IntelligencE' ,"Can computers think?" is specified to the question ,"Can computers formulate and justify hypotheses?" Various attempts have been made to answer the latter question positively. The present book is one such attempt. Our aim is not to formalize and mechanize the whole domain of inductive reasoning. Our ultimate question is: Can computers formulate and justify scientific hypotheses? Can they comprehend empirical data and process them rationally, using the apparatus of modern mathematical logic and statistics to try to produce a rational image of the observed empirical world? Theories of hypothesis formation are sometimes called logics of discovery. Plotkin divides a logic of discovery into a logic of induction: studying the notion of justification of a hypothesis, and a logic of suggestion: studying methods of suggesting reasonable hypotheses. We use this division for the organization of the present book: Chapter I is introductory and explains the subject of our logic of discovery. The rest falls into two parts: Part A - a logic of induction, and Part B - a logic of suggestion.

The book provides a comprehensive treatment of combinatorial development of heterogeneous catalysts. In particular, two computer-aided approaches that have played a key role in combinatorial catalysis and high-throughput experimentation during the last decade OCo evolutionary optimization and artificial neural networks OCo are described. The book is unique in that it describes evolutionary optimization in a broader context of methods of searching for optimal catalytic materials, including statistical design of experiments, as well as presents neural networks in a broader context of data analysis. It is the first book that demystifies the attractiveness of artificial neural networks, explaining its rational fundamental OCo their universal approximation capability. At the same time, it shows the limitations of that capability and describes two methods for how it can be improved. The book is also the first that presents two other important topics pertaining to evolutionary optimization and artificial neural networks: automatic generating of problem-tailored genetic algorithms, and tuning evolutionary algorithms with neural networks. Both are not only theoretically explained, but also well illustrated through detailed case studies. Sample Chapter(s). Chapter 1: Background of Combinatorial Catalyst Development (63 KB). Contents: Background of Combinatorial Catalyst Development (M Baerns); Approaches in the Development of Heterogeneous Catalysts (M Baerns); Mathematical Methods of Searching for Optimal Catalytic Materials (M Holena); Generating Problem-Tailored Genetic Algorithms for Catalyst Search (M Holena); Analysis and Mining of Data Collected in Catalytic Experiments (M Holena); Artificial Neural Networks in the Development of Catalytic Materials (M Holena); Tunning Evolutionary Algorithms with Artificial Neural Networks (M Holena); Improving Neural Network Approximations (M Holena); Applications of Combinatorial Catalyst Development and An Outlook on Future Work (M Baerns). Readership: Chemists and chemical engineers from academia and industry working in catalysis; materials scientists; graduate students dealing with catalytic chemistry interested in computer-aided methods.