Author: Vincent Diepeveen
Date: 06:55:11 01/31/00
Go up one level in this thread
>I am not sure. My company does ordering of 2^700 possibilities and photo/image >(tyres) recognition which also works by comparing patterns and so on. That works >fine so far, so why shouldn't it be possible to use at least some of these >techniques? It seems to me that you should talk to someone more intensively who >has a much knowlegde of fuzzy programming and fuzzy programming techniques. > >(No, not me, I have no idea how to program such methods, there are others who >can explain that to you much better than I can, and their explanations sound >senseful to me.) I think you should read very carefully what i write now: Recognizing voice and tyres isn't simple. It's very complicated. However, what you basically try to achieve is to recognize a known goal. The goal is a specific tyre, or a specific voice. Neural networks and similar approaches can be TRAINED to learn a specific voice, tyre or whatever. The real problem of these approaches is there when you apply them to something that isn't recognizing something known but searching in the unknown. If you train an neural network or whatever to recognize the openingsposition in chess, then i'm sure that in time the thing will learn what openingsposition is. The problem however in chess is that you need metaknowledge about a position which decides what you think from an UNKNOWN position. Basically you try to get an evaluation from something unknown, and here lies the problem: In chess you train for the unknown (positions), but in tyresrecognition you know EXACTLY for what tyre you are looking for, so it has a CLEAR and VERY EXACT goal. Now the goal in chess is putting mate, but that can be expressed already in a much simpler way. The same is valid for patterns. Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.