X hits on this document

PDF document

On pattern, categories, and alternate realities - page 10 / 14

40 views

0 shares

0 downloads

0 comments

10 / 14

Volume 14, Number 3

PATTERN RECOGNITION LETTERS

March 1993

an Australian aboriginal language, Dyirbal-but not on the basis that women are fiery and dangerous. Anything connected with water, the sun, and the stars is also placed in this category.

Alternate realities

One of the themes of the above mentioned books is a readiness to accept the existence of alternate perceptions of reality with alternate meanings and objectives. Consider the following. Sometimes, as in the goblet/two faces picture or drawings such as the young girl/old lady picture (Figure 5), or Wittgenstein's duck/ rabbit drawing, there are actually two different perceptions of reality simultaneously

Figure 5. Young woman/old woman drawing

represented by the same data. Other times, even if there be one reality known to an all-seeing God, it would in- evitably be reported differently by observers based on their individual perceptions. For example, there is the classic Japanese movie, Roshomon, in which different witnesses to a crime give very different accounts of the events that transpired. That such multiple perceptions of reality are quite common in scientific affairs is brought home to us by the following observations concerning the book Perceptrons written by Marvin Minsky and Seymour Papert. In (Hecht-Nielsen (1990)), Robert Hecht-Nielsen quotes from the book Perceptrons; a part of the extract is presented here:

"However, we were now involved in establishing at M.I.T. an artificial intelligence laboratory largely devoted to real `seeing machines,' and gave no attention whatsoever to perceptrons until we were jolted by attending an I.E.E.E. workshop on Pattern Recognition in Puerto Rico early in 1967. Appalled at the persistent influence of perceptrons (and similar ways of thinking) on practical pattern recognition, we determined to set out our work as a book..."

I mentioned this I.E.E.E. workshop at the beginning of this talk. I was one of the organizers and also edited the book Pattern Recognition (Kanal (1968)) which was the Proceedings of the workshop. As noted in the preface of the book, the workshop was held on October 24-26, 1966. As I have mentioned elsewhere, many workers had abandoned perceptron type learning machines and moved on to statistical and syntactic procedures. So only one session of five papers, chaired by Prof. Widrow of Stanford, was devoted to papers on adaptive networks for pattern recognition. Thus of the thirty papers in the book only five (5) are on this topic. Yet Minsky and Papert found themselves appalled at the persistent influence of perceptrons! I believe I attended most of the sessions. I had a different perception regarding the influence of perceptrons at the workshop. Whatever the reasons, their perception was different and motivated them to do some very interesting research on the single layer perceptron. I certainly would not say their book was the main reason for the demise of perceptron type networks. I think the main reason was inadequate technology and training algorithms for multilayer perceptrons and the unrealistic,

250

Document info
Document views40
Page views40
Page last viewedWed Dec 07 16:42:38 UTC 2016
Pages14
Paragraphs205
Words8189

Comments