,
4 tweets,
2 min read
Read on Twitter
@MatMcGann It starts gradually. First the lowest order stabilizes, then the second lowest, etc.
Or to be more precise:
t=0 everything unstable
t=1 first level half stable
t=2 first level stable, second level half stable
t=3 first two levels stable, third level half stable
And so on
Or to be more precise:
t=0 everything unstable
t=1 first level half stable
t=2 first level stable, second level half stable
t=3 first two levels stable, third level half stable
And so on
@MatMcGann Example: first, babies cannot even see.
Once they see, they can begin recognizing faces.
Once they recognize faces, they can begin recognizing expressions.
Once they recognize expressions, they can begin recognizing emotions in others.
And so on
Once they see, they can begin recognizing faces.
Once they recognize faces, they can begin recognizing expressions.
Once they recognize expressions, they can begin recognizing emotions in others.
And so on
@MatMcGann (Of course, not so black and white. Once they somehow learned to see faces, they can begin recognizing expressions. At the same time they’re learning a level of representation and perfecting all the ones below.)
@MatMcGann I’m mostly referring to human learning, but I think it should also apply to AI learning, in particular to systems which are not simple ML but which aim to learn in multiple contexts and without supervision (but the last part is just a conjecture of mine).
