🚀 go-pugleaf

(Mark Robert Thorson) wrote suggesting that a neural net should watch
program behaviour in order to learn how to manage the cache for that
program.

I wrote
> This is a joke, right?  I mean, neural nets typically require THOUSANDS
> of training runs to learn even the simplest things.  In this case, each
> training run corresponds to a complete execution of the program.  ...
rather grossly understating my case.

In article <22211@cup.portal.com>, mmm@cup.portal.com (Mark Robert Thorson) writes:
> The speed of learning depends on the model being used.  Would you have
> agreed with my statement if I had used the term "advanced statistical
> techniques" rather than the emotion-laden "neural networks"?

H*** no!  If you had said and meant "utterly simple methods justified by
advanced statistical arguments" I might have agreed.  Since Thorson
proposed no model, I assumed models like the ones in the PDP book, which
I have tried.

Fact 1:  there is nothing that a neural net can do that a parallel boolean
         circuit of similar complexity cannot do.  (Don't forget that the
         complexity of the coefficients has to be accounted for.)  If you
         want to call parallel boolean circuits "neural nets", fine.
Fact 2:  a system which can represent information has to be at least as
	 complex as the information it represents.
Fact 3:  the kind of information Thorson envisaged his hypothetical nets
	 learning is very complex:  this variable ought to be cached NOW
	 even though it shouldn't have been cached a few seconds ago.
Fact 4:  it follows from facts 2 and 3 that a neural net, parallel boolean
	 circuit, advanced statistical technique, or thaumatron system
	 for Thorson's caching scheme must be complex.
One is left wondering whether it might not be a better use of resources
to just spend the effort on a bigger cache.

[PS: "thaumatron" is a joke.
 It means an electronic device for performing miracles.]