Hi Aymeric
On Mon, Jan 3, 2022 at 1:08 PM aymeric <[email protected]> wrote:
>
>
> Hello everyone,
>
> I am defining my Phd subject which will revolve around Neuro-symbolim. The
> goal would be to make an agent to learn as much knowledge as it can from its
> environement, and I found the opencog and the atomspace graph very
> interesting for the implementation.
>
> My questions might be too vague, but maybe somebody can provide an answer or
> at least some guiding ideas.
In principle, your questions are answered in various wiki pages and
demos, but I don't recall if we have a single simple approachable wii
page for this. Nil, do you know of anything?
> How can we use the opencog system to store increasingly more complex concepts
> of the environment an agent is exploring ? Can the concepts be casted as
> atoms ? How would the relations between atoms be constructed ?
You have many choices in how to do this. A common and popular way to
do this is to write
(EvaluationLink
(PredicateNode "relative who is an uncle")
(ListLink
(ConceptNode "Bob")
(ConceptNode "Alice")))
which encodes that Bob is Alice's uncle. There are ways to attach
numbers to the above, to signify propabilities, certainties, weights,
and so on. You can attach key-value pairs to the five atoms above, as
you wish.
You don't even have to use EvaluationLinks or PredicateNodes or
ConceptNodes as above; you can represent this data in other ways,
depending on how clever and fancy you want to get. But the above is a
good place to start.
> Is there already a solution in the opencog ecosystem to store concepts in a
> hierarchical or composed manner ?
Sure. It's called the "AtomSpace". You can read about it here:
https://github.com/opencog/atomspace
The above points to additional documentation and demos.
> (I am thinking for instance of the Option framework,
I don't know what the "Option framework" is, and neither does
duckduckgo or google.
> where long sequences of actions are stored (as RL algos) and guided by a
> meta-policy,
I don't know what an "RL algo" is, or a "meta-policy".
> can this framework be "easily" implemented with opencog?).
OpenCog consists of dozens of pieces-parts. Some are strong, solid and
mature; some are broken and obsolete, and some are brand new and in
active development.
The AtomSpace is by far the most mature, stable and well-tested
component. It's the foundation stone for everything else. There are
other components you might be interested in, but you'd have to
describe what you want to do in greater detail.
-- Linas
--
You received this message because you are subscribed to the Google Groups
"opencog" 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/opencog/CAHrUA37odcZ%2B4JpPkYtN3a_PypOXBcdn3DY-Svvjfsi-mum8eQ%40mail.gmail.com.
