Using Genetic Programming to Extract Knowledge from Artificial Neural Networks

A technique is described in this chapter that makes it possible to extract the knowledge held by previously trained artificial neural networks. This makes it possible for them to be used in a number of areas (such as medicine) where it is necessary to know how they work, as well as having a network that functions. This chapter explains how to carry out this process to extract knowledge, defined as rules. Special emphasis is placed on extracting knowledge from recurrent neural networks, in particular when applied in predicting time series.

Application of Fuzzy Optimization in Forecasting and Planning of Construction Industry

This chapter proposes a new method to obtain optimal solution using satisfactory approach in uncertain environment. The optimal solution is obtained by using possibilistic linear programming approach and intelligent computing by MATLAB?. The optimal solution for profit function, index quality and worker satisfaction index in construction industry is considered. Decision maker and implementer tabulate the final possibilistic and realistic outcome for objective functions respect to level of satisfaction and vagueness for forecasting and planning. When the decision maker finds the optimum parameters with acceptable degree of satisfaction, he/she can apply the confidence of gaining much profit in terms of helping the public with high quality and least cost products. The proposed fuzzy membership function allows the implementer to find a better arrangement for the equipments in the production line to fulfill the wanted products in an optimum way.

Multi-Vehicle Missions

This chapter deals with the issues associated with the autonomy of vehicle fleets, as well as some of the dimensions provided by an Artificial Intelligence (AI) solution. This presentation is developed using the example of a suppression of enemy air defense mission carried out by a group of Unmanned Combat Air Vehicles (UCAV). The environment of the Mission Management System (MMS) includes the theatre of operations, vehicle sub-systems and the MMS of other UCAV. An MMS architecture, organized around a database, including reactive and deliberative layers is described in detail. The deliberative layer includes a distributed mission planner developed using constraint programming and an agent framework. Experimental results demonstrate that the MMS is able, in a bounded time, to carry out missions, to activate the contingent behaviors, to decide whether to plan or not. Some research directions remain open in this application domain of AI.

Induction as a Search Procedure

This chapter introduces Inductive Logic Programming (ILP) from the perspective of search algorithms in Computer Science. It first briefly considers the Version Spaces approach to induction, and then focuses on Inductive Logic Programming: from its formal definition and main techniques and strategies, to priors used to restrict the search space and optimized sequential, parallel, and stochastic algorithms. The authors hope that this presentation of the theory and applications of Inductive Logic Programming will help the reader understand the theoretical underpinnings of ILP, and also provide a helpful overview of the State-of-the-Art in the domain.

Parallelizing Genetic Algorithms

Genetic Algorithms (GA), which are based on the idea of optimizing by simulating the natural processes of evolution, have proven successful in solving complex problems that are not easily solved through conventional methods. This chapter introduces their major steps, operators, theoretical foundations, and problems. A parallel GA is an extension of the classical GA that takes advantage of a GA’s inherent parallelism to improve its time performance and reduce the likelihood of premature convergence. An overview of different models for parallelizing GAs is presented along with a discussion of their main advantages and disadvantages. A case study: A parallel GA for finding Ramsey Numbers is then presented. According to Ramsey Theory, a sufficiently large system (no matter how random) will always contain highly organized subsystems. The role of Ramsey numbers is to quantify some of these existential theorems. Finding Ramsey numbers has proven to be a very difficult task that has led researchers to experiment with different methods of accomplishing this task. The objective of the case study is both to illustrate the typical process of GA development and to verify the superior performance of parallel GAs in solving some of the problems (e.g., premature convergence) of traditional GAs.

Rank Improvement Optimization Using PROMETHEE and Trigonometric Differential Evolution

This chapter investigates the modelling of the ability to improve the rank position of an alternative in relation to those of its competitors. PROMETHEE is one such technique for ranking alternatives based on their criteria values. In conjunction with the evolutionary algorithm Trigonometric Differential Evolution, the minimum changes necessary to the criteria values of an alternative are investigated, for it to achieve an improved rank position. This investigation is compounded with a comparison of the differing effects of two considered objective functions that measure the previously mentioned minimization. Two data sets are considered, the first concerns the ranking of environmental projects, and the second the ranking of brands of a food product. The notion of modelling preference ranks of alternatives and the subsequent improvement of alternative’s rank positions is the realism of a stakeholders’ appreciation of their alternative in relation to their competitors.

Reinforcement Learning and Automated Planning

This article presents a detailed survey on Artificial Intelligent approaches, that combine Reinforcement Learning and Automated Planning. There is a close relationship between those two areas as they both deal with the process of guiding an agent, situated in a dynamic environment, in order to achieve a set of predefined goals. Therefore, it is straightforward to integrate learning and planning, in a single guiding mechanism and there have been many approaches in this direction during the past years. The approaches are organized and presented according to various characteristics, as the used planning mechanism or the reinforcement learning algorithm.

Optimising Object Classification

This chapter investigates the effectiveness of a number of objective functions used in conjunction with a novel technique to optimise the classification of objects based on a number of characteristic values, which may or may not be missing. The classification and ranking belief simplex (CaRBS) technique is based on Dempster-Shafer theory and, hence, operates in the presence of ignorance. The objective functions considered minimise the level of ambiguity and/or ignorance in the classification of companies to being either failed or not-failed. Further results are found when an incomplete version of the original data set is considered. The findings in this chapter demonstrate how techniques such as CaRBS, which operate in an uncertain reasoning based environment, offer a novel approach to object classification problem solving.

Single- and Multi-order Neurons for recursive unsupervised learning

In this chapter we present a recursive approach to unsupervised learning. The algorithm proposed, while similar to ensemble clustering, does not need to execute several clustering algorithms and find consensus between them. On the contrary, grouping is done between two subsets of data at one time, thereby saving training time. Also, only two kinds of clustering algorithms are used in creating the recursive clustering ensemble, as opposed to the multitude of clusterers required by ensemble clusterers. In this chapter a recursive clusterer is proposed for both single and multi order neural networks. Empirical results show as much as 50% improvement in clustering accuracy when compared to benchmark clustering algorithms.

Stratified Constraint Satisfaction Networks in Synergetic Multi-Agent Simulations of Language Evolution

We describe a simulation model of language evolution which integrates synergetic linguistics with multiagent modelling. On the one hand, this enables the utilization of knowledge about the distribution of the parameter values of system variables as a touch stone of simulation validity. On the other hand, it accounts for synergetic interdependencies of microscopic system variables and macroscopic order parameters. This approach goes beyond the classical setting of synergetic linguistics by grounding processes of selfregulation and self-organization in mechanisms of (dialogically aligned) language learning. Consequently, the simulation model includes four layers, (i) the level of single information processing agents which are (ii) dialogically aligned in communication processes enslaved (iii) by the social system in which the agents participate and whose countless communication events shape (iv) the corresponding language system. In summary, the present chapter is basically conceptual. It outlines a simulation model which bridges between different levels of language modelling kept apart in contemporary simulation models. This model relates to artificial cognition systems in the sense that it may be implemented to endow an artificial agent community in order to perform distributed processes of meaning constitution.