opencog-dev team mailing list archive

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

On Wed, Jun 18, 2008 at 11:29 AM, Ben Goertzel <ben@xxxxxxxxxxxx> wrote:
> Well, I'm not sure it needs to be explicitly considered...
>
> I mean, STI has reality for system dynamic because it affects rent ...
>
> The approach I'm suggesting now, using a Markov matrix, is basically
> a reaction-diffusion process, right?

Makes sense.

> The basic STI update equation (including a rent term that differs for
> Atoms in and out of the AF) is the reaction ... and the Markov-matrix-based
> spreading is the diffusion...
>
> At very least, the  Markov matrix approach seems easy to understand...
> so it should be easy to fix any problems that arise

Agree.

On an implementation detail - if I'm going to be generating Inverse
matrices, has anyone had to do anything similar in Novamente in the
past? i.e. is there existing code, or should I implement it from
scratch?

Apparently Newton's method (
http://en.wikipedia.org/wiki/Newton's_method ) can be use inverses of
earlier matrices to seed generating inverses of later matrices.
Although I'd still need to implement Gauss–Jordan elimination for the
first iteration.

Ben, is the idea of operations on matrices at all related to proposal
you had for a process of simple operations making changes to the
AtomSpace? If so, is it worth including some existing Matrix library
into OpenCog?

J