Athanasios K. Tsadiras

Comparing the inference capabilities of binary, trivalent and sigmoid fuzzy cognitive maps

In this paper, we compare the inference capabilities of three different types of fuzzy cognitive maps (FCMs). A fuzzy cognitive map is a recurrent artificial neural network that creates models as collections of concepts/neurons and the various causal relations that exist between these concepts/neurons. In the paper, a variety of industry/engineering FCM applications is presented. The three different types of FCMs that we study and compare are the binary, the trivalent and the sigmoid FCM, each of them using the corresponding transfer function for their neurons/concepts. Predictions are made by viewing dynamically the consequences of the various imposed scenarios. The prediction making capabilities are examined and presented. Conclusions are drawn concerning the use of the three types of FCMs for making predictions. Guidance is given, in order FCM users to choose the most suitable type of FCM, according to (a) the nature of the problem, (b) the required representation capabilities of the problem and (c) the level of inference required by the case.

Cognitive mapping and certainty neuron fuzzy cognitive maps

Cognitive maps (CMs) and fuzzy cognitive maps (FCMs) are well-established techniques that attempt to emulate the cognitive process of human experts on specific domains by creating causal models as signed/weighted directed graphs of concepts and the various causal relationships that exist between the concepts. They are mainly used for decision making and prediction. A number of extensions are proposed to increase their inference and representation capabilities. Certainty neuron fuzzy cognitive maps (CNFCMs) are proposed by the authors. This structure can be considered as a recurrent neural network with certainty neurons to be used that are neurons using a special kind of transfer function of two variables. The new transfer function employs the certainty factor handling function that was used in the MYCIN expert system, and also imposes a decay mechanism. In CMs and most FCMs, the activation level of every concept of the model, in a crisp on/off manner, can take one value among the two allowed values, −1 or 1. CNFCM allows the activation level to be any decimal in the interval [−1,1] increasing the representation capabilities of the model. The equations that are applied at the equilibrium points of the CNFCM are found. Through simulations, the dynamical behavior of CNFCMs is presented and the inference capabilities are illustrated in comparison to that of the classical FCM by means of an example.

The MYCIN certainty factor handling function as uninorm operator and its use as a threshold function in artificial neurons

Abstract The uninorm operator class as defined by Yager, unifies the t-norm and t-conorm operator classes and allows special kind of aggregation that depends on the identity element. In the paper, the MYCIN certainty factor handling function f M is proved to belong to the uninorm operator class. The shape of function f M is studied and its resemblance to artificial neurons threshold functions is established. The two variable function f M can be seen as an extension of typical neuron threshold functions to the three-dimensional space. The use of one of the two variables as a parameter gives the neuron tuning capabilities. A specific class of Artificial Neural Networks that cope with uncertainty and can use M as threshold function is also proposed.

An experimental study of the dynamics of the certainty neuron fuzzy cognitive maps

Abstract Certainty Neurons have been introduced as a new type of artificial neurons that use a two variable transfer function that provides them with memory capabilities and decay mechanism. They are used in fuzzy cognitive maps which is an artificial neural network structure that creates models as collections of concepts – neurons and the various causal relationships – weighted arcs that exist between them. An experimental study of the certainty neuron fuzzy cognitive maps (CNFCMs) dynamical behaviour is presented as this appears through simulations. Two control parameters are used: the symmetry of the systems weight matrix and the strength of the decay mechanism. The values of these two parameters can lead the system to exhibit stable fixed point behaviour, limit cycle behaviour or to collapse. The ways that the two control parameters cause the change of the systems dynamical behaviour from fixed point to limit cycle are also presented. The areas where the systems exhibit specific dynamical behaviour are identified.

Using fuzzy cognitive maps as a decision support system for political decisions

In this paper we use Fuzzy Cognitive Maps (FCMs), a well-established Artificial Intelligence technique that incorporates ideas from Artificial Neural Networks and Fuzzy Logic, to create a dynamic model of the Former Yugoslavian Republic of Macedonia (FYROM) Crisis in March 2001. FCMs create models as collections of concepts and the various causal relations that exist between these concepts. The decision capabilities of the FCM structure are examined and presented using a model that is developed based on the beliefs of a domain expert. The model is first examined statically using graph theory techniques to identify the vicious or the virtuous cycles of the decision process. The model is also tested dynamically thought simulations, in order to support political analysts and decision makers to their political decisions concerning the crisis. Scenarios are introduced and predictions are made by viewing dynamically the consequences of the corresponding actions.

A DSS for the buffer allocation of production lines based on a comparative evaluation of a set of search algorithms

In the design of production lines, the classical approach to the buffer allocation problem (BAP) is to use a search algorithm in association with an evaluative algorithm to obtain the mathematical optimum of the specified objective function. In practice, a choice often has to be made regarding which search algorithm to use for the efficient solution of the BAP. This paper gives the results of a carefully selected set of experiments on short (K = number of stations = 3, 4, , 11 stations), medium length (K = 12, 13, , 30 stations) and long lines (K = 40, 50, , 100 stations) and, within each line, small N (N = total number of buffer slots = K/2 if K is even; = (K – 1)/2 if K is odd), medium N (N = K + 1) and large N (N = 2K) to evaluate the effectiveness of the following five search algorithms: simulated annealing, genetic, tabu search, myopic and complete enumeration (where possible). The production lines are balanced and the single exponential machine at each station is perfectly reliable. All the experiments were run on a readily available desktop PC with the following specifications: Windows XP Professional Version 2002 Service Pack 3, Pentium® Dual-Core CPU E5300@2.60 GHz, 2.00 GB RAM. The measures of performance used are CPU time required and closeness to the maximum throughput achieved. The five search algorithms are ranked in respect to these two measures and certain findings regarding their performance over the experimental set are noted. The distributions of buffer slots to storage areas for the algorithm(s) leading to maximum throughput are examined and certain patterns are found, leading to indications for design rules. Based on the results of the above experiments, two additional sets of experiments were carried out, one using the simulated annealing algorithm for production lines of K = 3 to 20 and N = 1 to 20 (accounting for a total number of 360 different production lines) and another using the myopic algorithm for production lines of K = 3 to 80 and N = 1 to 120 (accounting for a total number of 9360 different production lines). These results may be used as references for comparison purposes in the international literature. Using the results from all sets of experiments, a decision support system (DSS) is designed and implemented which, as is illustrated, may assist production line designers in making decisions regarding the most appropriate of the five search algorithms tested to use for the BAP-A (the dual problem) and the BAP-B (the primal problem) for a wide class of production lines (consisting of K = 3 to 80 and N = 1 to 120).

An artificial neural network based decision support system for solving the buffer allocation problem in reliable production lines

One of the major design problems in the context of manufacturing systems is the well-known Buffer Allocation Problem (BAP). This problem arises from the cost involved in terms of space requirements on the production floor and the need to keep in mind the decoupling impact of buffers in increasing the throughput of the line. Production line designers often need to solve the Buffer Allocation Problem (BAP), but this can be difficult, especially for large production lines, because the task is currently highly time consuming. Designers would be interested in a tool that would rapidly provide the solution to the BAP, even if only a near optimal solution is found, especially when they have to make their decisions at an operational level (e.g. hours). For decisions at a strategic level (e.g. years), such a tool would provide preliminary results that would be useful, before attempting to find the optimal solution with a specific search algorithm. The aim of this study is to create such a tool. More specifically, an Artificial Neural Network (ANN) based decision support system is developed to assist production line designers in making decisions concerning the Buffer Allocation Problem (BAP) in reliable production lines. The aim of the ANN is to predict the performance of the production line based on its characteristics. The decision support system has been designed to allow for these data to be outputted in a user friendly format. To develop such an ANN, a large number of training and test data is required. To collect these data, extensive experiments were performed on a carefully chosen set of production lines. Because of its speed, the myopic algorithm was used as the search algorithm for the experiments. The performance of the ANN is examined for test sets of production lines and an average accuracy close to 99% is found. The performance of the ANN is compared with that of other well established surface fitting methods and its superiority is confirmed. Based on the results from (a) the experiments and (b) the developed ANN, a decision support system, called BAPANN, is designed and implemented. BAPANNs functionalities and capabilities are demonstrated via the use of illustrative scenarios, showing the effectiveness of the proposed method measured in terms of the required CPU time. In summary, BAPANN provides the production line designer with a powerful, efficient and accurate tool to make decisions on the buffer allocation problem for balanced reliable production lines. This is done in a convenient fashion without involving the designer in tedious and complex mathematical analysis.

Using fuzzy cognitive maps as a decision support system for political decisions: the case of Turkey’s integration into the European union

In this paper we use Fuzzy Cognitive Maps (FCMs), a well-established Artificial Intelligence technique that incorporates ideas from Artificial Neural Networks and Fuzzy Logic, to create a dynamic model of Turkey’s course towards its integration into the European Union, after the decision of December 18, 2004, according to which in October 3, 2005, Turkey will start negotiating its access to the European Union. FCMs create models as collections of concepts and the various causal relations that exist between these concepts. The decision capabilities of the FCM structure are examined and presented using a model developed based on the beliefs of a domain expert in the political situation in Turkey & European Union. The model is examined both statically using graph theory techniques and dynamically through simulations. Scenarios are introduced and predictions are made by viewing dynamically the consequences of the corresponding actions.

RuleML representation and simulation of Fuzzy Cognitive Maps

Fuzzy Cognitive Map (FCM) technique is a combination of Fuzzy Logic and Artificial Neural Networks that is extensively used by experts and scientists of a diversity of disciplines, for strategic planning, decision making and predictions. A standardized representation of FCMs accompanied by a system that would assist decision makers to simulate their own developed Fuzzy Cognitive Maps would be highly appreciated by them, and would help the dissemination of FCMs. In this paper, (a) a RuleML representation of FCM is proposed and (b) a system is designed and implemented in Prolog programming language to assist experts to simulate their own FCMs. This system returns results in valid RuleML syntax, making them readily available to other cooperative systems. The representation capabilities and the design choices of the implemented system are discussed and a variety of examples are given to demonstrate the use of the system.

A Rule Based Personalized Location Information System for the Semantic Web

In this paper, an innovative Personalized Location Information System for the Semantic Web (called SPLIS) is presented. The proposed system adopts schema.org ontology and combines it with rule-based policies, to deliver fully contextualized information to the user of a location-based system. Owners of points of interest can add their own rule-based policies to SPLIS to expose and deploy their marketing strategy on special offers, discounts, etc. These rules are combined at run-time with information about relevant place properties and user (people) profiles. Additionally, owners of points of interest can extend the ontology by adding dynamically specific properties. Rules are encoded in RuleML for interchangeability and to Jess in order to be executed. All data and rules are stored in the form of triples, using Sesame. Rules are evaluated on-the-fly to deliver personalized information according to the rules that fired within the current user-location-time context. In the paper, a demonstration of SPLIS is given using data from Google Places API and Google map for visualization.