X hits on this document

PDF document

On pattern, categories, and alternate realities - page 12 / 14





12 / 14

Volume 14, Number 3


March 1993

neuroscience. Artificial Intelligence and Pattern Recognition were viewed as a common enterprise. Part of the excitement was due to neural networks, especially perceptrons. After two decades of growth and splitting into narrower and narrower subfields, with many specialized names, there is once again a confluence of ideas and results from neuroscience, physics, psychology, computer science, engineering, mathematics, etc., and an appreciation of the need to have a more integrated or holistic approach to pattern recognition. Although the current round of excitement also started with neural nets, in practical applications hybrid methods, combining numeric, symbolic, statistical and structural, crisp and fuzzy representations and logics, are rapidly taking over center stage. The fast hardware, desk-top workstations with dazzling color displays, easy to use software tools and user-friendly interfaces all add to the renewed excitement about our work. But the main reason is that with the new hardware and software technology, and three decades of algorithm development, satisfactory solutions to some complex pattern recognition problems are being obtained.

Formalisms for developing algorithms and parallel implementations for many of the individual tools shown in Figure 1 have received much attention in recent years. But the integration of heterogeneous computational components, multiple sensors producing different types of data, and heterogeneous knowledge bases, is a significant systems design problem for which we currently have only ad-hoc techniques. Clearly more systematic methods and formalisms need to be developed for the design of complex multilevel systems consisting of hetero- geneous modules performing specialized local computations while interacting with other modules at the same and different levels of a hierarchical organization. Such interaction involves information and decisions flowing back and forth, with competition, and cooperation, all in the context of global constraint satisfaction. The management of hybrid systems, not unlike the management of complex human organizations, requires significant attention to language and communication between different modules for reasoning and decision making (Kanal and Perlis (1989)). In the process of developing formalisms for hybrid systems, foundations underlying current approaches should be open to question. This occurred, for example, with Lotfi Zadeh's development of fuzzy set theory; that theory should become ‘fuzzier’ with interval based computation (Womg.et al. (1992)). It is likely that the ability to model complex cognitive tasks using digital systems will be more and more, in question. In Kanal and Tsao (1987) we have expressed some doubts regarding perception being modelable by a Turing machine. The need for and possibility of non-Turing machines are likely to be topics of increasing discussion and argument in the coming years.

Concluding remarks

In its broadest sense pattern recognition is at the heart of all scientific inquiry, including our attempt to understand ourselves and the world around us. That task remains a most challenging one, but- what could be more interesting? In the IAPR we should continue to take a broad interdisciplinary view of pattern recognition, be open to new ideas, keeping in mind the multicultural nature of our enterprise.

And now it appears that the time has gone. But here's an alternate reality:

Le temps s'en va, le temps s'en va, madame! Las! le temps non: mais Nous nous en allons!

Time goes, you say? Ah, no! Alas Time stays, we go;

Pierre de Ronsard, The Paradox of Time (Austin Dobson, tr.)

This is a beautiful poem but ultimately a sad one and I don't want to end on a sad note. So I end with a final definition of "pattern". A "pattern" in Ireland is a feast or merriment in honor of a patron saint. Let us dedicate our banquet to the memory of King-Sun Fu. Let us have a "pattern" 3


Akaike, H. (1974). A new look at the statistical model identification. IEEE Trans. Automat. Control 19 (6), 716-723. Bahl, L.P. de Souza et al. (1992). A fast match for continuous speech using allophonic models. Proc. Internal. Conf. on


The banquet two nights later was indeed a feast of food and music.


Document info
Document views18
Page views18
Page last viewedThu Oct 20 20:02:58 UTC 2016