ROCCA Monthly Community Call

[Nil Geisweiller]

Hi everybody,

this is a reminder of the ROCCA monthly community call.

Time: Friday Sept 3 at 3pm EEST
Place: https://meet.jit.si/rocca

See you there!

Nil

[Adrian Borucki]

Here is the recording from the meeting: https://odysee.com/@ntoxeg:8/rocca_call_2021_09:7

It seems that the video has better frame rate than when recorded by Nil via OBS so if you want I can volunteer to record these via Jitsi each meeting from now on.

Cheers,

Adrian

[Nil Geisweiller]

Thanks Adrian!

Regarding local recording it looks like the jitsi community is working on it

https://github.com/jitsi/jitsi-meet/issues/6014

in the meantime there's workaround

https://github.com/TALRACE/JitsiLocalScreenRecorder

which I might try, just to give us extra options.

Nil

> Place: https://meet.jit.si/rocca <https://meet.jit.si/rocca>
>
> See you there!
>
> 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 view this discussion on the web visit
> https://groups.google.com/d/msgid/opencog/04888d56-4fc1-40c9-8a04-8c8341a82688n%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/04888d56-4fc1-40c9-8a04-8c8341a82688n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Adrian Borucki]

Just to make sure — is ROCCA call confirmed for tomorrow (Nov 5)?

[Nil Geisweiller]

Hi Adrian,

On 11/4/21 5:32 PM, Adrian Borucki wrote:
> Just to make sure — is ROCCA call confirmed for tomorrow (Nov 5)?

Thanks for the reminder! I've been quite busy (still am) and ROCCA is
in a broken state right now (being reworked to handle temporal
deduction) but we could still have a call nevertheless, what do you think?

Time would be 3pm EET (we're back to winter time) and place would be
https://meet.jit.si/rocca

Nil

> <https://groups.google.com/d/msgid/opencog/04888d56-4fc1-40c9-8a04-8c8341a82688n%40googlegroups.com?utm_medium=email&utm_source=footer
> <https://groups.google.com/d/msgid/opencog/04888d56-4fc1-40c9-8a04-8c8341a82688n%40googlegroups.com?utm_medium=email&utm_source=footer>>.

>
>
> --
> 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 view this discussion on the web visit

> https://groups.google.com/d/msgid/opencog/5d7b1985-7d16-461e-a949-c89828041fd7n%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/5d7b1985-7d16-461e-a949-c89828041fd7n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Adrian Borucki]

We could have a call, unless this is a burden to you, then we can postpone that to some other date.

I don’t have anything particular to talk about besides hearing some ROCCA updates (what is its state and roadmap at this point) and maybe pondering again about using MOSES for learning a useful representation of observations for ROCCA.

So, I am fine whether we have that call or not — the choice is yours, really.

Time and place is fine.

[Nil Geisweiller]

I have to leave right now, I'll be back in 30min, then we can have a call.

Nil

On 11/4/21 9:11 PM, Adrian Borucki wrote:
> We could have a call, unless this is a burden to you, then we can
> postpone that to some other date.
> I don’t have anything particular to talk about besides hearing some
> ROCCA updates (what is its state and roadmap at this point) and maybe
> pondering again about using MOSES for learning a useful representation
> of observations for ROCCA.
> So, I am fine whether we have that call or not — the choice is yours,
> really.
>
> Time and place is fine.
>
> On Thursday, 4 November 2021 at 18:48:01 UTC+1 Nil wrote:
>
> Hi Adrian,
>
> On 11/4/21 5:32 PM, Adrian Borucki wrote:
> > Just to make sure — is ROCCA call confirmed for tomorrow (Nov 5)?
>
> Thanks for the reminder! I've been quite busy (still am) and ROCCA is
> in a broken state right now (being reworked to handle temporal
> deduction) but we could still have a call nevertheless, what do you
> think?
>
> Time would be 3pm EET (we're back to winter time) and place would be

> https://meet.jit.si/rocca <https://meet.jit.si/rocca>

> <https://groups.google.com/d/msgid/opencog/5d7b1985-7d16-461e-a949-c89828041fd7n%40googlegroups.com?utm_medium=email&utm_source=footer
> <https://groups.google.com/d/msgid/opencog/5d7b1985-7d16-461e-a949-c89828041fd7n%40googlegroups.com?utm_medium=email&utm_source=footer>>.

>
>
> --
> 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 view this discussion on the web visit

> https://groups.google.com/d/msgid/opencog/755d85cd-639b-49b0-a703-c391e38c9effn%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/755d85cd-639b-49b0-a703-c391e38c9effn%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Adrian Borucki]

All right.

[Adrian Borucki]

Recording from the November meeting: https://odysee.com/@ntoxeg:8/rocca_call_2021_11:8?r=4wtHaEnW6caN3sFeQn3bhZGNUv56jY2s

[Nil Geisweiller]

Thanks Adrian.

I've created a wikipage about ROCCA and added links to our previous calls.

https://wiki.opencog.org/w/ROCCA#ROCCA_Community_Monthly_Call

Nil

> <https://groups.google.com/d/msgid/opencog/755d85cd-639b-49b0-a703-c391e38c9effn%40googlegroups.com?utm_medium=email&utm_source=footer
> <https://groups.google.com/d/msgid/opencog/755d85cd-639b-49b0-a703-c391e38c9effn%40googlegroups.com?utm_medium=email&utm_source=footer>>.

>
>
> --
> 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 view this discussion on the web visit

> https://groups.google.com/d/msgid/opencog/66693d70-b15e-476c-aef8-f3422207635an%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/66693d70-b15e-476c-aef8-f3422207635an%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Patrick Hammer]

Hi Nil!

Sorry I missed the November and October meetings, but I will make sure to attend in December.

I watched your AGI conference workshop presentation in the meanwhile: your work on temporal deduction is awesome, and I can't wait to see the PLN-based learner in action!

If you want I can show you some structures and control policies which might help you in case controlling the inference process (for induction and deduction) is a challenge.

I think it would also be interesting to explore the use of PLN induction rules via URE instead of relying on Pattern Miner.

In principle, finding temporal patterns can directly utilize the order of events happening in time and use the order of events for a strategy to control induction, so might not need the more general machinery.

But at the same time a PatternMiner+AtomSpace combination seems more general, which at some point might be better than relying on a custom control strategy operating with SpaceTime server.

Just some things to consider... :)

Best regards,

Patrick

[Nil Geisweiller]

Hi Patrick!

On 11/30/21 20:03, Patrick Hammer wrote:
> If you want I can show you some structures and control policies which
> might help you in case controlling the inference process (for induction
> and deduction) is a challenge.

Sure, I've been letting aside the inference control problem as getting
sound reasoning has been the challenge so far but it surely is a crucial
aspect.

> I think it would also be interesting to explore the use of PLN induction
> rules via URE instead of relying on Pattern Miner.

Yeah, I think ultimately the pattern miner should be framed more as a
PLN process, which is the case in theory already but not so much in
practice (partly for technical and efficiency reasons).

> In principle, finding temporal patterns can directly utilize the order
> of events happening in time and use the order of events for a strategy
> to control induction, so might not need the more general machinery.
> But at the same time a PatternMiner+AtomSpace combination seems more
> general, which at some point might be better than relying on a custom
> control strategy operating with SpaceTime server.
> Just some things to consider... :)

I think OpenCog and NARS differ a bit here. OpenCog has more like a
separated Attention Allocation mechanism working in parallel, while
NARS' is more integrated. That's an area I'm lacking experience (I feel
I'm pretty much blank slate on the subject). Happy to discuss it during
the call.

Best,
Nil

> <https://groups.google.com/d/msgid/opencog/66693d70-b15e-476c-aef8-f3422207635an%40googlegroups.com?utm_medium=email&utm_source=footer
> <https://groups.google.com/d/msgid/opencog/66693d70-b15e-476c-aef8-f3422207635an%40googlegroups.com?utm_medium=email&utm_source=footer>>.

>
>
> --
> 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 view this discussion on the web visit

> https://groups.google.com/d/msgid/opencog/723bce80-5f84-4381-b921-289c56ba13d7n%40googlegroups.com
> <https://groups.google.com/d/msgid/opencog/723bce80-5f84-4381-b921-289c56ba13d7n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Linas Vepstas]

On Thu, Dec 2, 2021 at 7:24 AM 'Nil Geisweiller' via opencog <ope...@googlegroups.com> wrote:

Hi Patrick!

I think OpenCog and NARS differ a bit here. [...] I'm pretty much blank slate on the subject

NARS in it's earliest conception was a paper by Pei Wang that described formulas for reasoning, not unlike the PLN formulas, but differing in details.  Part of the vision for URE was to allow NARS (or other kinds of deductive/inductive systems) to be implemented, as "just another set of rules".

I think this is still the case today. You could "easily" implement NARS today, by cloning the PLN repo, gutting 95% of the rules in it,  keeping around a couple as examples, and then start adding the NARS rules. It should be "easy" with an appropriate definition of "easy".

Now Ben didn't like NARS, back in the day, because some of the rules appeared to violate conventional Bayesian probability, and he figured his bread was buttered on the side closer to conventional probability theory. Thus PLN. But anyway, with the URE, you can experiment with any kind of deductive system you want.

These days, I don't like NARS or PLN, because I think I know how to learn common-sense reasoning "from scratch". That includes automatically learning PLN-like or NARS-like rules, whatever they may be, as appropriate to whatever setting they've been learned for.  Doing this is on my TODO list, but  at the current rate of progress, this is at least a decade away.

--linas

--

Patrick: Are they laughing at us?

Sponge Bob: No, Patrick, they are laughing next to us.

 

[Nil Geisweiller]

On 12/2/21 18:09, Linas Vepstas wrote:
> NARS in it's earliest conception was a paper by Pei Wang that described
> formulas for reasoning, not unlike the PLN formulas, but differing in
> details.  Part of the vision for URE was to allow NARS (or other kinds
> of deductive/inductive systems) to be implemented, as "just another set
> of rules".

NARS is more than that, what you are referring to is NAL, Non-Axiomatic
Logic.

> I think this is still the case today. You could "easily" implement NARS
> today, by cloning the PLN repo, gutting 95% of the rules in it,  keeping
> around a couple as examples, and then start adding the NARS rules. It
> should be "easy" with an appropriate definition of "easy".

True, in fact I've considered doing it, to get a one to one comparison.
Not high on the priority list but very interesting nonetheless.

> These days, I don't like NARS or PLN, because I think I know how to
> learn common-sense reasoning "from scratch". That includes automatically
> learning PLN-like or NARS-like rules, whatever they may be, as
> appropriate to whatever setting they've been learned for.  Doing this is
> on my TODO list, but  at the current rate of progress, this is at least
> a decade away.

Interesting. I guess though in the end you still need a metric of
comparison.

Sometimes I wonder if NAL isn't actually based on probability theory
with a very unconventional prior.

Nil

>
> --linas
>
> --
> Patrick: Are they laughing at us?
> Sponge Bob: No, Patrick, they are laughing next to us.
>
>

> --
> 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 view this discussion on the web visit

> https://groups.google.com/d/msgid/opencog/CAHrUA356X0upDHw%3DdfXsA%3DYLc_FeTwTHRUrqF6QFrE3P-kV8iQ%40mail.gmail.com
> <https://groups.google.com/d/msgid/opencog/CAHrUA356X0upDHw%3DdfXsA%3DYLc_FeTwTHRUrqF6QFrE3P-kV8iQ%40mail.gmail.com?utm_medium=email&utm_source=footer>.

[Linas Vepstas]

On Thu, Dec 2, 2021 at 11:52 PM Nil Geisweiller <ngei...@googlemail.com> wrote:

> These days, I don't like NARS or PLN, because I think I know how to
> learn common-sense reasoning "from scratch". That includes automatically
> learning PLN-like or NARS-like rules, whatever they may be, as
> appropriate to whatever setting they've been learned for.  Doing this is
> on my TODO list, but  at the current rate of progress, this is at least
> a decade away.

Interesting. 

I gave a talk on that at AGI 2021.  My todo list is to convert it to a blog post.

I guess though in the end you still need a metric of
comparison.

I don't mean to discourage you. I would still like to see a faster URE/PLN and also an easier-to-use/comprehend URE/PLN. I'd like to see more users, so the easier-to-use part gets higher priority.

In the past, I've glimpsed how a certain subset of openpsi is a lot like the URE, and I'd like to see that explored some more. It would be interesting if both could be made to have very similar or the same API, allowing comparisons.

As long as it is just you and I (and occasionally Ben) in these conversations, it remains hard to pursue these ideas.

Sometimes I wonder if NAL isn't actually based on probability theory
with a very unconventional prior.

Huh. Maybe? The axioms of probability theory are very generic; it would indeed be hard to evade them. Perhaps the underlying set on which NAL acts is not what it seems to be.

--linas