Interested in contributing

120 views
Skip to first unread message

James Bolden

unread,
Apr 17, 2019, 11:30:52 AM4/17/19
to opencog
Hi,

My name is James Bolden. I'm an upperclassman math and philosophy student with significant computer science experience. 
My philosophy background has been concerned primarily with philosophy of mind and consciousness studies. My experience in 
mathematics has been aimed towards abstract algebra and mathematical logic. Anyway, I'm interested in contributing to OpenCog
in whatever capacity I can. However, I'm not totally sure where to start. I've done a lot of reading on the wiki and played around 
some with the atomspace. But I've only got a basic grasp of some of the key systems (this project is big). Any pointers or direction 
would be greatly appreciated.

Thanks,
James

Nil Geisweiller

unread,
Apr 17, 2019, 11:43:03 AM4/17/19
to ope...@googlegroups.com
Hi James,

your help on PLN, and logic/reasoning in general could be greatly
appreciated!

OpenCog has an rewriting system, call the Unified Rule Engine (URE),
akin to Prolog, with some special operators to communicate with the
AtomSpace (OpenCog's hypergraph data base) and manipulate sub-symbolic
knowledge, probabilities, tensors, etc.

On top of the URE is implemented PLN (Probabilistic Logic Network), but
really any type of logic can be implemented, even higher order (albeit
with some inconveniences since the unifer is first-order-ish).

I suggest you go straight to the examples

https://github.com/opencog/atomspace/tree/master/examples/ure

start with simple, then frog.

Then go to the PLN examples if you wish

https://github.com/opencog/opencog/tree/master/examples/pln

start with good-song.

There's plenty more stuff to explore. For instance we have a pattern
miner that runs on top the URE, a program learner called MOSES, etc.

Nil
> --
> 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 opencog+u...@googlegroups.com
> <mailto:opencog+u...@googlegroups.com>.
> To post to this group, send email to ope...@googlegroups.com
> <mailto:ope...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/5d11793d-9327-413c-8f11-df7560bb72a0%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/5d11793d-9327-413c-8f11-df7560bb72a0%40googlegroups.com?utm_medium=email&utm_source=footer>.
> For more options, visit https://groups.google.com/d/optout.

James Bolden

unread,
Apr 17, 2019, 12:02:13 PM4/17/19
to opencog
Thank you for your prompt response and pointers, Nil. Both URE and PLN look to be in my wheelhouse.

I'm reading through simple and frog at the moment. It looks like rewrite rules are written in Atomese. 
I assume this is because the goal is for most, if not all, rules to be learned rather than built-in. How 
is rule learning accomplished, i.e., what subsystem would have dominion over that process?

Nil Geisweiller

unread,
Apr 17, 2019, 1:16:24 PM4/17/19
to ope...@googlegroups.com
On 4/17/19 7:02 PM, James Bolden wrote:
> I'm reading through simple and frog at the moment. It looks like rewrite
> rules are written in Atomese.
> I assume this is because the goal is for most, if not all, rules to be
> learned rather than built-in. How

Correct. :-)

> is rule learning accomplished, i.e., what subsystem would have dominion
> over that process?

If we're talking about logical rules, ultimately I suppose they should
be reasoned out, at least if we want to guaranty that they are correct.
Outside of that, any thing that can spit a BindLink should do the trick,
such as MOSES (once fully ported to support Atomese) or the pattern miner.

Orthogonally you may want to have a look at

https://github.com/opencog/opencog/tree/master/examples/pln/inference-control-learning

a little experiment to learn Control Rules (as opposed to Inference
Rules) to guide reasoning, using the pattern miner as learner.

Nil
> > an email to ope...@googlegroups.com <javascript:>
> > <mailto:ope...@googlegroups.com <javascript:>>.
> > To post to this group, send email to ope...@googlegroups.com
> <javascript:>
> > <mailto:ope...@googlegroups.com <javascript:>>.
> <https://groups.google.com/group/opencog>.
> <https://groups.google.com/d/msgid/opencog/5d11793d-9327-413c-8f11-df7560bb72a0%40googlegroups.com?utm_medium=email&utm_source=footer
> <https://groups.google.com/d/optout>.
>
> --
> 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 opencog+u...@googlegroups.com
> <mailto:opencog+u...@googlegroups.com>.
> To post to this group, send email to ope...@googlegroups.com
> <mailto:ope...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/7a5e7b0a-e4b0-4f8a-aa1a-a70d3b436352%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/7a5e7b0a-e4b0-4f8a-aa1a-a70d3b436352%40googlegroups.com?utm_medium=email&utm_source=footer>.

Linas Vepstas

unread,
Apr 18, 2019, 6:08:27 PM4/18/19
to opencog
On Wed, Apr 17, 2019 at 12:16 PM 'Nil Geisweiller' via opencog <ope...@googlegroups.com> wrote:
On 4/17/19 7:02 PM, James Bolden wrote:
>
> is rule learning accomplished, i.e., what subsystem would have dominion
> over that process?

If we're talking about logical rules, ultimately I suppose they should
be reasoned out, at least if we want to guaranty that they are correct.
 
I have vague, general ideas about how they can be learned statistically, but there are various preliminary steps that need to be accomplished first.  The meaning of the word "correct" is a very interesting philosophical question. (viz 10-year-olds reason "correctly", despite not having a firm foundation in Aristotelian logic. three-fourths-of-a-century of Piaget has elucidated more; the trick is to convert real-life situations into observational statistics for unsupervised machine learning.)

--linas

--
cassette tapes - analog TV - film cameras - you
Message has been deleted

Linas Vepstas

unread,
Apr 22, 2019, 8:25:34 PM4/22/19
to opencog


On Mon, Apr 22, 2019 at 7:17 PM Sarah Weaver <lwfl...@gmail.com> wrote:
So the mailing list is the main way the community operates these days?

Yes. There is also an IRC channel, but no one uses it. There are multiple slack channels, that no one uses. I think slack is broken-by-design, and would prefer if matrix/riot was used instead. But since one one uses any of these, it doesn't much matter, I guess.


I'm interested, although I may need to learn better methods: my general approach is an "Artificial Environment" and not Artificial Intelligence ... per say. And that may well be the entirely wrong thing.

Opencog has many different parts, some more mature than others; some experimental, some obsolete.  I don't know what to recommend, it depends on what you are trying to do.

--linas 

Sarah Weaver

unread,
Apr 23, 2019, 9:05:15 PM4/23/19
to ope...@googlegroups.com
For right now, I'm going for something between Narrow AI and AGI. If anyone remembers science classes and evolution, I sometimes jokingly call it the Australopithecus stage. Basic fight or flight response fine tuned, but it's still primitive enough where certain granular safety measures are programmed.

Biggest challenge is learning other machine learning besides Decision Trees. Certain gems (if I still use Ruby) I can't seem to install.

--
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 opencog+u...@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
Reply all
Reply to author
Forward
Message has been deleted
0 new messages