Re: [Algorithms] Intelligent 'Snap' behaviour in tools
Brought to you by:
vexxed72
From: Anders N. <ja...@an...> - 2006-07-12 00:46:13
|
So just build a nerual-net and train it! Give it just enough context and start with a simple snapping method. Every time snapping is performed, the user result of that snapping is evaluated. "Ok we snapped X to Y, then he changed it and it ended up in Z, and then he finally got it to place W. But we would have gotten it there anyway given method R." This is a way to dynamically build the decision tree. Nice thing is that it could adapt to different users, devices etc. Hard thing is to determine correct context (especially if it's a hiarchy) (and all the usual NNET headaches) If the NNET is trained in-house it could be used to generate the decision-tree; just feed it with the context and see what it does for each decision. I don't really think it's practical just throwing it into the debate. Every time I hear the word expert system I think; neural net! :) Anders Nilsson. |