¥ protect human life
¥ humans society is thr greatest threat to human life
... ... ... ... ...
... ... ... ... ...
¥ remove human legs, take away eyes
how does one improve decision tree making methods though?
Well, one thing is making a more stable splitting criterion. Right now the existing decision tree algorithms (which were invented in fucking
eighties and they're still being fucking used - it's unbelievable just how stale the field is) basically splits information into two roughly equal halves based on maximizing information gain, which means it's very unstable to small changes in database - especially if there are several attributes which are equally good for purpose of splitting. Fixing that would go a long way.
Thats part of the top down approach to AI. Its been attempted for about 30 years or something now, and it kinda failed miserably :u
Today research is focusing on the bottom-up approach, IE creating something that can perceive things and make decisions based on pattern recognition, which is how actual animals (and us) do things.
That makes no sense to me. Like, literally no sense. Neural network algorithms are not fundamentally different from decision trees at all - you feed some known labeled data into both of them, they each process it, and produce a fixed classifier at the end of their operation, which is then used to classify unknown examples. Hell, they are usually mentioned next to each other in all statistical classification books that I've read as equally valid ways of doing things, with each having their own advantages and disadvantages.
What do you mean by this "top-down vs bottom-up" thing? A genuine question here.