INTERNATIONAL SYMPOSIUM on INTELLIGENT SYSTEMS AMSE-ISIS'97

UNIVERSITY OF REGGIO CALABRIA
FACULTY OF ENGINEERING

REGGIO CALABRIA, ITALY

September 11-13, 1997

PLENARY TALKS:
Prof. Dr. Jaime Gil-Aluja
"The Theory Of Affinities In The Economy And The Management"

Abstract:
The social, economic and managerial systems are enduring deeper changes, unknown in the past, as a consequence of the speed and force with which they appear.
This implies some effects amongst we stand out: the mutability of the phenomena and the uncertainty of future events.
From a time, scientists have intended to find elements, as well theoretical as technical, able to explain and to treat the new economic and managerial realities.
Between the phenomenas, which with most frequency appear in the economic-financial context of our days, stand out the forming of sets and subsets, of both forma and material beings which possess certain common characteristics. The need to classify the groups constitutes one of the most pressing problems in the contemporary economic systems.
The solutions to those problems were, not long-time ago, through the obtention of similarity relations "represented" by symmetric and reflexive matrices, almost ever from making the complement to the distance. Adequate algorithms gave the possibility of obtaining maximum subrelations of similarity. It was and important step but it was unenable to completely solve the problem since this way was just for square matrices.
Some years ago, Prof. Arnold Kaufmann and I, have started to attempt at the generalisation of the similarity concept with the object to approach the treatment of relations represented through rectangular matrices. As a result of this works we got what we call "theory of affinities"
The work we present for this occasion pretends to put on relief the fundamental elements which configure the concept of affinity as well as the demonstration that the subrelations of the obtained "covering" form a Galois lattice. Its use in the field of the economy and the management have permitted to bring solution to problems which until now would not be treat in all they extension.


Prof. Harold Szu
"Neural Network Adaptive Wavelet Transforms (AWT) for signal and pattern recognitions"

Abstract:
Super-Mother Theorem for AWT (Szu & Telfer, Opt. Eng. Sept. 94) will show why in principle AWT is better than the popular Radial Base Function NN Approach. This is because of the convergence proof of the Theorem, a much general class of zero-mean logic base wavelet functions than the radial base functions can be used to approximate the desired functional behaviors in terms of the super-mother wavelets which are capable of adapting itself to a very general class of bipolar analog signals of zero mean. Applications to variety cases will be given.


Prof. Simon Haykin
"Chaos, Radar Clutter, and Neural Networks"

Abstract:
In this lecture, I will cover three topics that are closely related, making a story that is both coherent and exciting. First, I will briefly describe the background theory of chaotic processes, with emphasis on the algorithms that should be followed to establish if a given time series is drawn from a chaotic process. Then, I will proceed to use these tools to demonstrate, in a conclusive way, that sea clutter (i.e., radar backscatter from an ocean surface) is a chaotic process. In effect, I will use real-life radar data for a case study. Finally, I will describe how to exploit the knowledge that sea clutter is systems.


Prof. Chris Oliver
"Information from SAR texture"

Abstract:
This review discusses the principles underlying methods for extracting information from natural clutter textures in SAR images. Initially we examine the role of the product model as a texture descriptor. In particular we compare various empirical PDFs with the gamma-distributed RCS model leading to a K-distributed intensity. This comparison justifies the use of the K-distribution to represent natural clutter. The remainder of the presentation is then concerned with exploiting this model to extract information.
We consider the impact of the model on three different types of processing function:
a) texture extimation,
b) classification,
c) segmentation.
In each case we examine the effect of the model on exploitation. Optimised texture estimation depends on identifying Maximum Likelihood (ML) estimators for the texture parameters. Classification operates by comparing data samples with references which can be analytic or result from training. Classification based on ML fit values for the parameter estimates is compared with direct ML classification. For single-point statistics the results are shown to be the same. However, for correlated textures, direct classification based on the amplitude spectral coefficients is preferable. Segmentation depends on exploiting the data alone to determine where the edges between regions of different texture occur. ML edge detection using the amplitude spectrum coefficients and that based on the single-point statistics are shown to give the same performance for uncorrelated textures. However, for correlated textures it is shown that edge detection based on the spectrum is less sensitive to changes in order than the method based on the PDF directly.
Throughout the results will be illustrated by simulation and a practical application to detecting clearings in the Amazon rain forest.


Prof. Boris Stilman
"Linguistic Geometry for Intelligent Systems"

Abstract:
There are many real-world problems where human expert skills in reasoning about complex systems are incomparably higher than the level of modern computing systems. At the same time there are even more areas where advances are required but human problem-solving skills can not be directly applied. For example, there are problems of planning and automatic control of autonomous agents such as space vehicles, stations, and robots with cooperative and opposing interests operating in a complex, hazardous environment. Reasoning about such complex systems should be done automatically, in a timely manner, and often in real time. Moreover, there are no highly-skilled human experts in these fields ready to substitute for robots (on a virtual model) or transfer their knowledge to them. There is no grand-master in robot control, although, of course, the knowledge of existing experts in this field should not be neglected - it is even more valuable. It is very important to study human expert reasoning about similar complex systems in the areas where the results are successful, in order to discover the keys to success, and then apply and adopt these keys to the new, as yet, unsolved problems. It should look like an application of the methods of a chess expert to a robot control or maintenance scheduling and vice versa. What mathematical tools can be applied to the area of successful results in order to generate a formal, domain independent knowledge ready to be transferred?
This talk is intended to introduce conference participants to the main ideas of research on geometrical properties of complex multiagent systems, Linguistic Geometry. This research relies on the formalization of search heuristics of highly-skilled human experts that have resulted in the development of successful applications in different areas. It is based on broad application of the theory of formal languages and grammars as well as theories of formal planning employing first order predicate calculus. The syntactic tools allow us to decompose complex concurrent multiagent system into the dynamic hierarchy of subsystems, a hierarchy of the networks of paths, and, thus, solve otherwise intractable problems by reducing the search dramatically (e.g., from trillions to tens of states). ... High Assurance Software, Emergency Vehicles Routing, Mobil Robot Control, Space Navigation, Combat Simulation and Control, etc. This is just a small subset of the long list of applications of Linguistic Geometry.
No preliminary knowledge is required.