opencog-dev team mailing list archive

["Trent Waddington" <trent.waddington@xxxxxxxxx>]

On Sat, Jun 21, 2008 at 3:21 AM, Andre Luiz de Senna
<senna@xxxxxxxxxxxxx> wrote:
> If your NN algorithm don't have benefits from possible interation
> with other Novamente entities (PLN etc) I don't see why you
> should use the AtomTable node hierarchy to represent your
> networks.
>
> I'm sorry if i missed some important discussion regarding this
> design. If that's the case please point me to the proper threads.

No, I'm just thinking out loud.

The power of NNs is almost completely a function of "how you wire them
up".  What you decide to put in the input layer determines what is
correlated.  What you duplicate in the input layer or carry over from
previous epochs greatly affects what the NN actually calculates.  And
then where you put your output layer or whether you use a generative
NN and pull your output from the recreated inputs, and how many of
them, etc, etc.

These are the kinds of decisions that I'd like to have the engine
make, so different kinds of NNs with different learning algorithms can
be run on internal or external data sets to find useful features.

Trent