Internal models of the environment have an important role to play in adaptive systems in general and are of particular importance for the supervised learning paradigm. In this paper we demonstrate that certain classical problems associated with the notion of the \teacher " in supervised learning can be solved by judicious use of learned internal models as components of the adaptive system. In particular, we show how supervised learning algorithms can be utilized in cases in which an unknown dynamical system intervenes between actions and desired outcomes. Our approach applies to any supervised learning algorithm that is capable of learning in multi-layer networks.

Abstract | We survey learning algorithms for recurrent neural networks with hidden units, and put the various techniques into a common framework. We discuss xedpoint learning algorithms, namely recurrent backpropagation and deterministic Boltzmann Machines, and non- xedpoint algorithms, namely backpropagation through time, Elman's history cuto, and Jordan's output feedback architecture. Forward propagation, an online technique that uses adjoint equations, and variations thereof, are also discussed. In many cases, the uni ed presentation leads to generalizations of various sorts. We discuss advantages and disadvantages of temporally continuous neural networks in contrast to clocked ones, continue with some \tricks of the trade" for training, using, and simulating continuous time and recurrent neural networks. We present somesimulations, and at the end, address issues of computational complexity and learning speed.

: This study investigates the properties of arti#cial recurrent neural networks. Particular attention is paid to the question of how these nets can be applied to the identi#cation and control of non-linear dynamic processes. Since these kind of processes can only insu#ciently be modelled by conventional methods, di#erent approaches are required. Neural networks are considered to be useful for this purpose due to their ability to approximate a wide class of continuous functions. Among the numerous network structures, especially the recurrentmulti-layer perceptron #RMLP# architecture is promising from application point of view. This network architecture has the wellknown properties of multi layer perceptrons and moreover these nets have the ability to incorporate temporal behavior. Departing from the original process description the applicability of RMLPs is investigated and di#erent learning algorithms for this network class are outlined. Furthermore, besides the conventional...

. A brief introduction to the main ideas in Autonomous Control Systems is first given and certain important issues in modeling, analysis and design are discussed. Control systems with high degree of autonomy should perform well under significant uncertainties in the system and environment for extended periods of time, and they must be able to compensate for certain system failures without external intervention. Highly autonomous control systems evolve from conventional control systems by adding intelligent components, and their development requires interdisciplinary research. A working characterization of intelligent controllers is introduced and it is argued that the supervisory controller discussed here, which can learn events, is indeed intelligent. There are problems in Autonomous Control Hybrid control systems are of great importance in the development of autonomous control and they are discussed extensively. An appropriate hybrid system model is first introduced and it is used to...

This paper deals with the problem of looper control for tension-free rolling. Conventional controllers cannot deal eectively with unmodeled dynamics and large variations which can lead to scrap runs and damages to machinery. Therefore, a fuzzy controller has been designed to use the expert knowledge of the operators for disturbedprocess control. Also, a self-tuning algorithm is incorporated for both on-line and o-line tuning of the fuzzy membership functions. This paper discusses the design of the fuzzy logic controller and its self-tuning. The eects of various design options are discussedand practical conclusions are made. Results from simulations arealsopresented. 1 Introduction The loop control methods are commonly used for exible cross-sections at the intermediate and nishing submills. These methods rely on an initial formation of a bar loop by utilizing mechanical deectors and proper motor speed adjustments (Fig. 1). In fact, eachstand roll speed has to be synchronized t...

Arti#cial neural networks are being used with increasing frequency for high dimensional problems of regression or classi#cation. This article provides a tutorial overview of neural networks, focusing on back propagation networks as a method for approximating nonlinear multivariable functions. We explain, from a statistician's vantage point, why neural networks might be attractive and how they compare to other modern regression techniques. KEYWORDS: nonparametric regression; function approximation; backpropagation. 1 Introduction Networks that mimic the way the brain works; computer programs that actually LEARN patterns; forecasting without having to know statistics. These are just some of the many claims and attractions of arti#cial neural networks. Neural networks #we will henceforth drop the term arti#cial, unless we need to distinguish them from biological neural networks# seem to be everywhere these days, and at least in their advertising, are able to do all that statistics...

This paper presents an application of multilayered feed-forward neural networks for controlling a PT326 Process Trainer laboratory equipment. Firstly, the process as well as its inverse have been identi#ed using the back-propagation #BP# algorithm for neural network training. Secondly an internal model control #IMC# strategy has been used for neurocontrol. Di#erent architectures and learning methods have been investigated for model approximation. Control of the process has been implemented in real-time using the Simulink#Matlab environment. Experimental results regarding the performance of the control scheme are included in a comparative study. 1. Introduction Recent progresses in control theory have made possible the developmentofadvanced control systems relying on model based control strategies. Due to the capabilities of non-linear function approximation with an arbitrary degree of accuracy, neural networks are an optimal tool to non-linear system modelling #1, 2#. Therefore...

The rich materials on modeling and control using linear system theory do not mean that the world is linear rather than nonlinear, but actually reflect our awkward situation of having too few mathematical tools available to deal with the complex nonlinear systems in reality. In many cases, the representation of signals and descriptions of systems are not necessarily the best but sacrificed for the convenience of mathematics. As the systems become ever complex, the drawbacks of linear system description become prominent such as its “local” applicability. Thanks to the collective efforts of many researchers, many significant and fundamental contributions have been made in neural network (NN)