Using Hierarchical Recurrent Neuro-Fuzzy Systems for Decision Making

AAAI Conferences

Over the last years, a number of methods have been proposed to automatically learn and optimize fuzzy rule bases from data. The obtained rule bases are usually robust and allow an interpretation even for data sets that contains imprecise or uncertain information. However, most of the proposed methods are still restricted to learn and/or optimize single layer feed-forward rule bases. The main disadvantages of this architecture are that the complexity of the rule base increases exponentially with the number of input and output variables and that the system is not able to store and reuse information of the past. Thus temporal dependencies have to be encoded in every data pattern. In this article we briefly discuss the advantages and disadvantages of hierarchical recurrent fuzzy systems that tackle these problems. Furthermore, we present a neuro-fuzzy model that has been designed to learn and optimize hierarchical recurrent fuzzy rule bases from data.

A Temporal Neuro-Fuzzy Monitoring System to Manufacturing Systems Artificial Intelligence

Fault diagnosis and failure prognosis are essential techniques in improving the safety of many manufacturing systems. Therefore, on-line fault detection and isolation is one of the most important tasks in safety-critical and intelligent control systems. Computational intelligence techniques are being investigated as extension of the traditional fault diagnosis methods. This paper discusses the Temporal Neuro-Fuzzy Systems (TNFS) fault diagnosis within an application study of a manufacturing system. The key issues of finding a suitable structure for detecting and isolating ten realistic actuator faults are described. Within this framework, data-processing interactive software of simulation baptized NEFDIAG (NEuro Fuzzy DIAGnosis) version 1.0 is developed. This software devoted primarily to creation, training and test of a classification Neuro-Fuzzy system of industrial process failures. NEFDIAG can be represented like a special type of fuzzy perceptron, with three layers used to classify patterns and failures. The system selected is the workshop of SCIMAT clinker, cement factory in Algeria.

Neuro-Fuzzy Computing System with the Capacity of Implementation on Memristor-Crossbar and Optimization-Free Hardware Training Artificial Intelligence

In this paper, first we present a new explanation for the relation between logical circuits and artificial neural networks, logical circuits and fuzzy logic, and artificial neural networks and fuzzy inference systems. Then, based on these results, we propose a new neuro-fuzzy computing system which can effectively be implemented on the memristor-crossbar structure. One important feature of the proposed system is that its hardware can directly be trained using the Hebbian learning rule and without the need to any optimization. The system also has a very good capability to deal with huge number of input-out training data without facing problems like overtraining.

A Neuro-Fuzzy Strategy for Web Personalization

AAAI Conferences

In this paper, we propose the use of a neuro-fuzzy strategy to develop a Web personalization framework for the dynamic suggestion of URLs retained interesting for the currently connected users. In particular, a hybrid strategy exploiting the combination of the fuzzy logic with the neural paradigm is proposed in order to discover useful knowledge from session data identified from the analysis of log files and represent it in a set of fuzzy rules expressed in an interpretable form.

Modular Neuro-Fuzzy Networks Used in Explicit and Implicit Knowledge Integration

AAAI Conferences

A framework of new unified neural and neuro-fuzzy approaches for integrating implicit and explicit knowledge in neuro-symbolic systems is proposed. In the developed hybrid system, training data set is used for building neurofuzzy modules, and represents implicit domain knowledge. On the other hand, the explicit domain knowledge is represented by fuzzy rules, which are directly mapped into equivalent neural structures. Three methods to combine the explicit and implicit knowledge modules are proposed.