@Plinz I just listened to a talk by Prof Gurdeep Hura of Univ of Maryland, ES.
He was asking: can machines do one work and then seemlessly jump to another work like we humans do multitasking.
Your thoughts on how to do it.
-
-
Replying to @muralipiyer
Similar to humans: learn a function that orchestrates lower level functions. The regulation of learned behaviors is itself a learned behavior.
1 reply 0 retweets 1 like -
Replying to @Plinz
Learn to use which algorithm to use? When context changes, how can a machine know that the context changed? We humans do a lot of mistakes ourselves in this.
1 reply 0 retweets 0 likes -
Replying to @muralipiyer
In the simplest case, you notice that your current model is not predictive any more, and you test which of your previous ones is best. You'll do better when you know the space of possible contexts, expect context changes or even plan them yourself.
1 reply 0 retweets 1 like -
Replying to @Plinz
What if we don't know all possible contexts? Say other person presumes that the system knows a certain topic and shifts conversation to that.
1 reply 0 retweets 0 likes -
Replying to @muralipiyer
You learn the escape probability (which tells you that none of your current models cut it) and start training a new model, usually by cloning a default model or the best current fit.
1 reply 1 retweet 0 likes -
Replying to @Plinz @muralipiyer
In an application/conversation context, only proceed when you have high certainty that you operate with a good model. As soon as you have reason to doubt that, stop and inform the user. This also holds when we are interacting with other people.
1 reply 1 retweet 0 likes -
Replying to @Plinz
This is why I love AI/ML...teaches us humans more than we manage to teach the machines!
1 reply 0 retweets 1 like
I see AI as a way to think about and understand ourselves!
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.