Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Yoink]

Funkahdafi. I cannot stop giggling.

That is pretty great.     

[Helgoland]

Funkahdafi. I cannot stop giggling.

That is pretty great.     

They even have a Malevich-inspired album cover.

[Sergarr]

It's ridiculous how little research there is into decision trees, given how well they (technically, methods derived on their basis, but that's details) perform, and given that they're about the only directly human-understandable (and human-constructible!) method that works well on non-linear problems (which is most of them).

I'm gonna try to make a prediction: one of the next breakthroughs in field of artificial intelligence & human-machine integration will come through the help of new & improved decision tree methods.

[pisskop]

¥  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?

[TempAcc]

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.

[pisskop]

thats fair, but isnt such a matrix more probe to errors?

its not like we have even identifyed all the 'patterns' we react to yet

[Frumple]

We'll know when we finish making them, presumably. Not that being prone to errors is really a problem, exactly. They just need to be less prone (on average) than current tech and/or humans.

[TempAcc]

Ye, people are still working out the whole pattern recognition thing. There's already stuff that can recognize patterns in words, images and sounds but thats still a long way from anything that could hope to pass a touring test, but its already worlds ahead of anything that used the top down approach (IE decision trees and etc).

[Sergarr]

¥  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.

[TempAcc]

Those are terms Ray Kurzweil used in one episode of exploration, which is this radio show thing hosted by Michio Kaku in which he interviews people and talks about weird science news thingies. I'll try to find it and post it here (fake edit: here), I have to go home so I'm not going to go into it myself, but I think they start talking about it after the 40 minute mark, or around then.

[Sergarr]

Those are terms Ray Kurzweil used in one episode of exploration, which is this radio show thing hosted by Michio Kaku in which he interviews people and talks about weird science news thingies. I'll try to find it and post it here (fake edit: here), I have to go home so I'm not going to go into it myself, but I think they start talking about it after the 40 minute mark, or around then.

Yeah, OK, I've listened to it, and decision trees are definitely the "bottom-up" approach. You seem to have it confused with the sort of a "rule-based" logical system, where you have a lot of "common-sense" and field-specific semantic/logical laws and rules and then use some sort of logic engine in order to produce statements and theories. Those did failed, I know that very well.

But decision trees are not like that at all. What they do is, in simple terms, repeatedly split given data on data attributes to put differently classified data into different places, and then use the resulting information construct to classify other data. It's data-based, in other words, just like neural algorithms are. But they're much less developed - last bit of serious research into core algorithm was made in 1984 - 32 years ago (CART).

I refuse to believe that after all these years, advancements in other fields have enabled no improvements to be made here.

[IronTomato]

https://www.youtube.com/watch?v=t-qReWLWOxs

[Levi]

Some people are actually trying to hack the internet. As far as hacking the internet can happen in reality, that is.

I should pre-download a few more of my GoG games, in case of a netpocalypse.

[Eric Blank]

Some people are actually trying to hack the internet. As far as hacking the internet can happen in reality, that is.

We are so screwed

[Neyvn]

Want my WTF for today...
Previously had a problem with Steam Autoupdating when I don't want it to due to Limited Net Allowance. Set time to do Autoupdates at a time I am not online, 12pm-1pm.
Come home at 4pm to chat and jobsearch.
Get Email that I am over allowance again.
Look at Steam. YEP DOWNLOADING A FUCKING UPDATE AGAIN!