opencog-dev team mailing list archive

["Joel Pitt" <joel.pitt@xxxxxxxxx>]

Hi Ben,

Today I've been implementing some different rent systems (as mentioned
as an idea in a previous email).

I've come up with an exponential and logarithmic based curve in
addition to the flat rent.

http://opencog.org/wiki/Image:Sti_rent_schemes.png

Ideally the area under the curve of each would be the same (I guess
you'd make the curve integrals equal and solve, but I also want to
make the y and x intercepts configurable as these act as a kind of
rent amnesty). However, I just wanted to get something working for the
moment...

Seems to make the Hopfield network work better as it allows those
atoms just in the AF to remain, and prevents some atoms become all
important, but could still be improved.

While drifting off to sleep last night I had a couple of ideas for the
importance spreading. Don't spread importance "uphill": atoms can't
spread their importance to an atom that is more important, and also
only spread enough to make their importance equivalent. Simple in
hindsight, but I think this, along with graduated rents, will make
things settle in the AF better.

J