Browse Prior Art Database

Introduction of Artificial Data to Enable Natural Selection Scoring Mechanism to Handle More Complex Cases

IP.com Disclosure Number: IPCOM000082566D
Original Publication Date: 1974-Dec-01
Included in the Prior Art Database: 2005-Feb-28
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Dunham, B: AUTHOR [+3]

Abstract

The concept of a design-by-natural-selection program is set forth in a paper "Design by Natural Selection", B. Dunham, et al, published in Syntheses, vol. 15, page 254, Sept. 1963. Such programs are typically applied to complex design problems where an optimized solution is sought. It is required that any particular solution can be scored.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 60% of the total text.

Page 1 of 1

Introduction of Artificial Data to Enable Natural Selection Scoring Mechanism to Handle More Complex Cases

The concept of a design-by-natural-selection program is set forth in a paper "Design by Natural Selection", B. Dunham, et al, published in Syntheses, vol. 15, page 254, Sept. 1963. Such programs are typically applied to complex design problems where an optimized solution is sought. It is required that any particular solution can be scored.

The procedure works roughly as follows. First, a random solution is generated and scored. Next, a randomly selected small change is tried. If a better score results, the change is accepted; otherwise, not. The process is continued, generating ever better solutions, until finally it becomes stuck in that no single change improves the score.

At this point, depending upon the application, a "traffic jam" procedure is introduced which enables the backing off slightly from the stuck position, and then to try to move forward again. The process is usually terminated after some suitable number of tries made without producing new progress.

When this basic method is adapted to real design-automation situations, many secondary criteria arise which can often be satisfied by the one basic "scorer" if the data is artificially modified to influence certain choices. For example, one important design automation program assigns circuits to one of two possible areas, minimizing the number of connections between the two areas. Suppose this pr...