Rough‐Neural Computing, Techniques for Computing with Words

Citation

Andrew, A.M. (2005), "Rough‐Neural Computing, Techniques for Computing with Words", Kybernetes, Vol. 34 No. 5, pp. 744-745. https://doi.org/10.1108/03684920510595445

Both of these introduce very new ideas in artificial intelligence mainly inspired by an initiative of Lotfi Zadeh, the originator of fuzzy set theory and hence of the revolution sparked by it. The second book starts with a foreword by Zadeh in which he defends the new viewpoint, arguing that the traditional emphasis on numerical data in science is unwarranted and that modelling and computing with words offer valuable new possibilities in which the primacy of numerical information is challenged.

Zadeh quotes Lord Kelvin as defending the primacy of numerical information, as a focus for his defence of the alternative. This is, I think, a little unfair since Kelvin was certainly enthusiastic about principles that could only be expressed verbally, but claimed that a test of their soundness was the possibility of numerical data. The same test is accepted today and even these new theories are judged by their performance in applications that either use numerical data or whose output is evaluated using numerical statistical criteria. (As a graduate in Natural Philosophy from Glasgow I am perhaps slightly partisan!).

Since the time of Kelvin, digital computation has developed in ways that even he could not have visualised. It is now possible to consider automated modelling and computing with words rather than numbers, and these new methods appear to substitute words for numbers in established procedures including adaptation in neural nets. The basic elements are information granules rather than neurons.

A distinction is drawn between modelling with words and computing with words, though Zadeh believes the gap will narrow. The combination of rough sets and neural networks, denoted by rough‐neural computing, fits the computing requirement, and the first of the books has a foreword and an introductory chapter by Professor Zdzislaw Pawlak, the founder of rough set theory. The book has 28 chapters by different authors, including a final part devoted to case studies.

One paper is by Witold Pedrycz, the authority on fuzzy system theory, and describes the use of the methods to allow operation of a distributed system of collaborative databases. Other papers refer to the aspects of audio and visual perception, including handwritten digit recognition, and to biomedical inference. Two others refer to aspects of audio perception, one of them to the acquisition of audio signals and the other to the influence of visual clues on surround sound perception. Others deal with signal classification, and with rule discovery in data. The robustness of the methods in the face of uncertainty and noise is emphasised throughout.

The topics of the 11 papers in the second book are on the whole rather more abstract, but the general thrust is the same, and it is asserted that these new techniques allow the integration of information from different and diverse sources, one of the advantages being the emergence of simpler learning algorithms because of better utilisation of past experience. An illustration on the cover shows a face with describable features picked out, implying an application to face recognition based on automatic linguistic description.

Both books are intended to be introductory, and the second is described by Zadeh as reader‐friendly. Nevertheless, the material in either is not accessible in any depth without some background. Even so it is impossible not to be impressed by the enthusiasm of the authors and to feel that this could be the beginning of something very big.