On 9/10/2019 1:10 PM, Philip Thrift wrote:
Deep nets are "algorithms" too. One can print out the gazillion
weights of the "neural" sigmoid functions of the connections
after it has deep-learned. That's just an algorithm that a human
couldn't read very well, because if it is printed out, would be
quite big.
And its reasoning is more like human intuition. In general, it
can't explain its process in a way that you could adopt it.
Brent
DeepNets also include modularity and interpretability:
@ Google Research
https://ai.googleblog.com/2019/09/recursive-sketches-for-modular-deep.html
https://ai.googleblog.com/2018/03/the-building-blocks-of-interpretability.html
So maybe they will report "explanations" soon. maybe better than
humans can their own.
From what I've read about DN the reporting of explanations is done by
additional nets and so it has the same problem as a human telling you
how to do something that they do intuitively (like hit a tennis
ball)...the explanation may not really align with what their brain does.
Brent
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/everything-list/be57689f-d533-7961-87a1-fd6677a007dc%40verizon.net.
