again we can set this up in the forum and also include voting.

2010/3/9 YKY (Yan King Yin, 甄景贤) <generic.intelligence@gmail.com>

> We need to have a little web applet for people to vote for the best BSer at
> the end of the session. I guess we can get that free somewhere online...
>
> YKY
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>



--
Carlos A Mejia

Taking life one singularity at a time.
www.Transalchemy.com
http://transalchemy.blogspot.com/

On 3/8/10, Edward Porter <ewporter@msn.com> wrote:

>
> ----Ed---->
> But it would appears just selecting which synapses activating which
> dendrites of neurons in a given volume could create the type of
> directionality we are talking about. Specifically what besides the changing
> of synapse weights is playing a role in directionality and how?

Yes these changes in AP’s directivity are likely to be correlated with
activities in synapses. Since the axon, dendrites are not simple
electric cables, changes at molecular level that occur with learning
have tremendous effect on AP propagation, selective activation of ion
channel ...
> -----Dorian----->

> ----Ed---->
> It would appears you are saying the expertise does not really exist in any
> one neuron (the weak learners), as much as it exists in a volume of neurons
> (the strong learner), so it would seem the term "expert" neuron is a
> misnomer. Is that correct?

Indeed, a strong expertise does not always exist in a single selected neuron.

> -----Dorian----->

>
> ----Ed---->
> What do you mean here? Do you mean that the neuron has a rhythmic or cyclic
> tendency that controls when synaptic transfers can cause an activation
> potential to propagate to their synapse before and during the transfer, and
> away from the synapse during or after the transfer?
>
> Wolf Singer implied that the rhythms of synchrony enable a neuron to
> selectively tune into signals from distant neurons that have the same
> synchrony. This implies synchrony greatly modulates the responsiveness of
> neurons to various axonic transfers that are attempting to be made to their
> dendrites. Is this type of synchrony part of your theory?
>

Yes, the neuron has a rhythmic behavior and processing information is
rhythmically performed in every neuron.
Indeed, Prof. Singer is right; synchrony increases the interaction and
implicitly boosts the computing power. Synchrony is a result of
interactions that occur at different levels including molecular level
where information is physically processed, from this perspective
synchrony is more like an epiphenomenon. This issue is explained in
detail in the book.


> -----Dorian----->

>
> ----Ed---->
> What is the relationship between large scale brain waves measured by EEG in
> each of the alpha, beta, theta, and gama bands?

The separation in delta, theta, alpha... is based on frequency
bandwidth and the existence of these rhythms reflect the overall
activity that occurs in neurons and synapses. Alterations in the
quantitative spectral characteristics of these rhythms are correlated
with information processing, transfer in different parts of the brain
and depend on the subject state (awake or sleep). It is likely that a
decrease/ increase of powers in some frequency bands is correlated
with processes that develop at molecular scale (gene selection,
appropriate protein synthesis…)

On Wed, Mar 10, 2010 at 4:50 AM, Matt Mahoney <matmahoney@yahoo.com> wrote:

> It might help if you define the problem you want to solve.
>
> AGI to pass the Turing test?
> AGI to launch a singularity?
> AGI for a better spam filter?
> AGI as a global brain to automate the world economy?
> AGI for data compression?
>

I was thinking of more specific sub-problems of AGI for each session, such
as:

1. how to make people with different AGI theories collaborate better?

2. how to make a distributive AGI with disparate components -- where they
don't even have to communicate via a fixed standard protocol or language?

3. how to avoid programming the AGI (or doing it as little as possible)

4. will there eventually be just one AGI or more than 1? If >1 , what
could be the possible features that differentiate them?

5. how to avoid the logic-centric AGI route that some of us are pursuing
now (eg OpenCog, NARS, Cyc, and my Genifer)?

etc etc....

YKY

We need to have a little web applet for people to vote for the best BSer at
the end of the session. I guess we can get that free somewhere online...

YKY

The biggest prize should go to the person who can define the problem!

On Tue, Mar 9, 2010 at 3:56 PM, The Wizard <key.universe@gmail.com> wrote:

> knowing the problem now would be cheating :P
>
>
> On Tue, Mar 9, 2010 at 3:50 PM, Matt Mahoney <matmahoney@yahoo.com> wrote:
>
>> It might help if you define the problem you want to solve.
>>
>> AGI to pass the Turing test?
>> AGI to launch a singularity?
>> AGI for a better spam filter?
>> AGI as a global brain to automate the world economy?
>> AGI for data compression?
>>
>>
>> -- Matt Mahoney, matmahoney@yahoo.com
>>
>>
>> *From:* "YKY (Yan King Yin, 甄景贤)" <generic.intelligence@gmail.com>
>>
>> *To:* agi <agi@v2.listbox.com>
>> *Sent:* Tue, March 9, 2010 2:35:00 PM
>>
>> *Subject:* Re: [agi] brainstorming contests?
>>
>> On Wed, Mar 10, 2010 at 3:15 AM, David Jones <davidhere40@gmail.com>wrote:
>>
>>> Hey YKY,
>>>
>>> I'm not sure. On one hand, ideas are a dime a dozen.
>>
>>
>> Good ideas are hard to come by and brainstorming definitely helps -- at
>> least, I think I'm very unoriginal and unspontaneous -- almost to the point
>> of autistic.
>>
>> It also helps that the good ideas are *shared* by a group instead of
>> belonging to and believed by just one person.
>>
>>
>>> On the other hand a prize and direction for the ideas could help create
>>> some useful ideas. The hard part is testing and implementing the idea well.
>>>
>>
>> We can try at least 2 sessions to see how it goes, one e-mail and one on
>> IRC. IRC is real-time and so may be more spontaneous, which might be better
>> for brainstorm.
>>
>> Maybe we can give SSD drives as prizes =)
>>
>> YKY
>> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
>> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
>> <http://www.listbox.com&gt;
>>
>> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
>> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
>> <http://www.listbox.com&gt;
>>
>
>
>
> --
> Carlos A Mejia
>
> Taking life one singularity at a time.
> www.Transalchemy.com
> http://transalchemy.blogspot.com/
>
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>



--
Brad Thomas
Founder
www.instansa.com
Twitter @bradleymthomas, @instansa
Linkedin www.bradleythomas.com

knowing the problem now would be cheating :P

On Tue, Mar 9, 2010 at 3:50 PM, Matt Mahoney <matmahoney@yahoo.com> wrote:

> It might help if you define the problem you want to solve.
>
> AGI to pass the Turing test?
> AGI to launch a singularity?
> AGI for a better spam filter?
> AGI as a global brain to automate the world economy?
> AGI for data compression?
>
>
> -- Matt Mahoney, matmahoney@yahoo.com
>
>
> *From:* "YKY (Yan King Yin, 甄景贤)" <generic.intelligence@gmail.com>
>
> *To:* agi <agi@v2.listbox.com>
> *Sent:* Tue, March 9, 2010 2:35:00 PM
>
> *Subject:* Re: [agi] brainstorming contests?
>
> On Wed, Mar 10, 2010 at 3:15 AM, David Jones <davidhere40@gmail.com>wrote:
>
>> Hey YKY,
>>
>> I'm not sure. On one hand, ideas are a dime a dozen.
>
>
> Good ideas are hard to come by and brainstorming definitely helps -- at
> least, I think I'm very unoriginal and unspontaneous -- almost to the point
> of autistic.
>
> It also helps that the good ideas are *shared* by a group instead of
> belonging to and believed by just one person.
>
>
>> On the other hand a prize and direction for the ideas could help create
>> some useful ideas. The hard part is testing and implementing the idea well.
>>
>
> We can try at least 2 sessions to see how it goes, one e-mail and one on
> IRC. IRC is real-time and so may be more spontaneous, which might be better
> for brainstorm.
>
> Maybe we can give SSD drives as prizes =)
>
> YKY
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>



--
Carlos A Mejia

Taking life one singularity at a time.
www.Transalchemy.com
http://transalchemy.blogspot.com/

It might help if you define the problem you want to solve.

AGI to pass the Turing test?
AGI to launch a singularity?
AGI for a better spam filter?
AGI as a global brain to automate the world economy?
AGI for data compression?

-- Matt Mahoney, matmahoney@yahoo.com


>
>From: "YKY (Yan King Yin, 甄景贤)" <generic.intelligence@gmail.com>
>To: agi <agi@v2.listbox.com>
>Sent: Tue, March 9, 2010 2:35:00 PM
>Subject: Re: [agi] brainstorming contests?
>
>
>On Wed, Mar 10, 2010 at 3:15 AM, David Jones <davidhere40@gmail.com> wrote:
>
>>>Hey YKY,
>>
>>I'm not sure. On one hand, ideas are a dime a dozen.
>
>
>Good ideas are hard to come by and brainstorming definitely helps -- at least, I think I'm very unoriginal and unspontaneous -- almost to the point of autistic.
>
>
>It also helps that the good ideas are *shared* by a group instead of belonging to and believed by just one person.
>
>>> On the other hand a prize and direction for the ideas could help create some useful ideas. The hard part is testing and implementing the idea well.
>>
>
>
>We can try at least 2 sessions to see how it goes, one e-mail and one on IRC. IRC is real-time and so may be more spontaneous, which might be better for brainstorm.
>
>
>Maybe we can give SSD drives as prizes =)
>
>
>YKY
>>
>agi | Archives > | Modify > Your Subscription

On Wed, Mar 10, 2010 at 3:15 AM, David Jones <davidhere40@gmail.com> wrote:

> Hey YKY,
>
> I'm not sure. On one hand, ideas are a dime a dozen.


Good ideas are hard to come by and brainstorming definitely helps -- at
least, I think I'm very unoriginal and unspontaneous -- almost to the point
of autistic.

It also helps that the good ideas are *shared* by a group instead of
belonging to and believed by just one person.


> On the other hand a prize and direction for the ideas could help create
> some useful ideas. The hard part is testing and implementing the idea well.
>

We can try at least 2 sessions to see how it goes, one e-mail and one on
IRC. IRC is real-time and so may be more spontaneous, which might be better
for brainstorm.

Maybe we can give SSD drives as prizes =)

YKY

Hey YKY,

I'm not sure. On one hand, ideas are a dime a dozen. On the other hand a
prize and direction for the ideas could help create some useful ideas. The
hard part is testing and implementing the idea well.



2010/3/9 YKY (Yan King Yin, 甄景贤) <generic.intelligence@gmail.com>

> On Tue, Mar 9, 2010 at 10:17 AM, Mike Tintner <tintner@blueyonder.co.uk>
> wrote:
> > In principle it's a v.g. idea - having dedicated brainstorming threads
> where
> > the recognized object is to come up with ideas, and toss ideas around
> freely
> > without people being too critical - but I don't think you have to
> > brainstorm in real time - if anything, it may be better if people have a
> > little time to respond to the latest idea, rather than having to respond
> > immediately. [Maybe you should mark such a thread with a prefix like
> > [Brainstorming] - or [BS] :) ]
>
> Some people who are regulars on the IRC channel may not be on this
> list... but we can test both e-mail brainstorms and IRC ones.
>
> We need to organize it as a contest and maybe get some sponsors (I'll
> see if David Jones is interested in sponsoring it, and I may also
> contribute some $).
>
> Also, the participants or some judges would need to vote for the best
> ideas.
>
> The topics of brainstorms are also open to suggestions each time...
>
> YKY
>
>
> -------------------------------------------
> agi
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/
> Modify Your Subscription:
> https://www.listbox.com/member/?&amp;
> Powered by Listbox: http://www.listbox.com
>

Off the cuff, I think brainstorming and contests are opposed - the former
IMO is supposed to be cooperative - but I'd be interested in further
thoughts. (Cooperative, see).

--------------------------------------------------
From: "YKY (Yan King Yin, ???)" <generic.intelligence@gmail.com>
Sent: Tuesday, March 09, 2010 6:23 AM
To: "agi" <agi@v2.listbox.com>
Subject: Re: [agi] brainstorming contests?

> On Tue, Mar 9, 2010 at 10:17 AM, Mike Tintner <tintner@blueyonder.co.uk>
> wrote:
>> In principle it's a v.g. idea - having dedicated brainstorming threads
>> where
>> the recognized object is to come up with ideas, and toss ideas around
>> freely
>> without people being too critical - but I don't think you have to
>> brainstorm in real time - if anything, it may be better if people have a
>> little time to respond to the latest idea, rather than having to respond
>> immediately. [Maybe you should mark such a thread with a prefix like
>> [Brainstorming] - or [BS] :) ]
>
> Some people who are regulars on the IRC channel may not be on this
> list... but we can test both e-mail brainstorms and IRC ones.
>
> We need to organize it as a contest and maybe get some sponsors (I'll
> see if David Jones is interested in sponsoring it, and I may also
> contribute some $).
>
> Also, the participants or some judges would need to vote for the best
> ideas.
>
> The topics of brainstorms are also open to suggestions each time...
>
> YKY
>
>
> -------------------------------------------
> agi
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/
> Modify Your Subscription:
> https://www.listbox.com/member/?&amp;
> Powered by Listbox: http://www.listbox.com

post it on the forum, although I do believe email or irc gets more dedicated
hits, yet I believe the forum just needs more dedicated posters

2010/3/9 YKY (Yan King Yin, 甄景贤) <generic.intelligence@gmail.com>

> On Tue, Mar 9, 2010 at 10:17 AM, Mike Tintner <tintner@blueyonder.co.uk>
> wrote:
> > In principle it's a v.g. idea - having dedicated brainstorming threads
> where
> > the recognized object is to come up with ideas, and toss ideas around
> freely
> > without people being too critical - but I don't think you have to
> > brainstorm in real time - if anything, it may be better if people have a
> > little time to respond to the latest idea, rather than having to respond
> > immediately. [Maybe you should mark such a thread with a prefix like
> > [Brainstorming] - or [BS] :) ]
>
> Some people who are regulars on the IRC channel may not be on this
> list... but we can test both e-mail brainstorms and IRC ones.
>
> We need to organize it as a contest and maybe get some sponsors (I'll
> see if David Jones is interested in sponsoring it, and I may also
> contribute some $).
>
> Also, the participants or some judges would need to vote for the best
> ideas.
>
> The topics of brainstorms are also open to suggestions each time...
>
> YKY
>
>
> -------------------------------------------
> agi
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/
> Modify Your Subscription:
> https://www.listbox.com/member/?&amp;
> Powered by Listbox: http://www.listbox.com
>



--
Carlos A Mejia

Taking life one singularity at a time.
www.Transalchemy.com
http://transalchemy.blogspot.com/

On Tue, Mar 9, 2010 at 10:17 AM, Mike Tintner <tintner@blueyonder.co.uk> wrote:
> In principle it's a v.g. idea - having dedicated brainstorming threads where
> the recognized object is to come up with ideas, and toss ideas around freely
> without people being too critical -  but I don't think you have to
> brainstorm in real time - if anything, it may be better if  people have a
> little time to respond to the latest idea, rather than having to respond
> immediately. [Maybe you should mark such a thread with a  prefix like
> [Brainstorming]   - or [BS]  :)  ]

Some people who are regulars on the IRC channel may not be on this
list... but we can test both e-mail brainstorms and IRC ones.

We need to organize it as a contest and maybe get some sponsors (I'll
see if David Jones is interested in sponsoring it, and I may also
contribute some $).

Also, the participants or some judges would need to vote for the best ideas.

The topics of brainstorms are also open to suggestions each time...

YKY

In principle it's a v.g. idea - having dedicated brainstorming threads where
the recognized object is to come up with ideas, and toss ideas around freely
without people being too critical - but I don't think you have to
brainstorm in real time - if anything, it may be better if people have a
little time to respond to the latest idea, rather than having to respond
immediately. [Maybe you should mark such a thread with a prefix like
[Brainstorming] - or [BS] :) ]

--------------------------------------------------
From: "YKY (Yan King Yin, ???)" <generic.intelligence@gmail.com>
Sent: Tuesday, March 09, 2010 1:13 AM
To: "agi" <agi@v2.listbox.com>
Subject: Re: [agi] brainstorming contests?

> On Tue, Mar 9, 2010 at 3:20 AM, Mike Tintner <tintner@blueyonder.co.uk>
> wrote:
>> If you'd like more contributions, why not post some of your brainstorming
>> on
>> here? - The more the channels of this group deversify, the more
>> difficult
>> it gets to follow - and the more likely it gets to break up altogether.
>
> If you guys prefer we can have brainstorm contests on this list, via
> e-mails. It's just that ordinarily brainstorms are carried out in
> real time, so I assume IRC is better for it, but we can definitely try
> e-mail brainstorms.
>
> If it's on IRC we would post the chat log here too.
>
> My suggestion is to hold low-cost contests and brainstorm seems to be
> easy to do. I can't think of other small project ideas... maybe
> someone else can?
>
> YKY
>
>
> -------------------------------------------
> agi
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/
> Modify Your Subscription:
> https://www.listbox.com/member/?&amp;
> Powered by Listbox: http://www.listbox.com

On Tue, Mar 9, 2010 at 3:20 AM, Mike Tintner <tintner@blueyonder.co.uk> wrote:
> If you'd like more contributions, why not post some of your brainstorming on
> here? -  The more the channels of this group deversify, the more difficult
> it gets to follow - and the more likely it gets to break up altogether.

If you guys prefer we can have brainstorm contests on this list, via
e-mails. It's just that ordinarily brainstorms are carried out in
real time, so I assume IRC is better for it, but we can definitely try
e-mail brainstorms.

If it's on IRC we would post the chat log here too.

My suggestion is to hold low-cost contests and brainstorm seems to be
easy to do. I can't think of other small project ideas... maybe
someone else can?

YKY

Steve Richfield wrote:
> Hence, importance is measured by compressibility as Matt indicated, but then again this is the SAME measure used to measure LACK of information content, and such input would be seen to carry LOW information content. Lower information content implies greater importance? Perhaps, but there must obviously be some scaling applied, e.g. measuring the "importance" by looking at how much it compresses to? Clearly, more thought is needed here. Matt, can you continue this thought?

Define the "importance" of X to Y as K(Y|X) - K(Y), where K() is Kolmogorov complexity. It is how much smaller the compressed size of Y can get when X is available for input.

K(Y|X) - K(Y) <= K(X), the information content of X. Some or all of the information in X is useful to predicting Y. Therefore for X to be important, it must have a lot of information.

However, having a lot of information does not imply importance. For example, suppose that X and Y are random and independent: K(X) = |X|, K(Y) = |Y|, K(X,Y) = K(X)+K(Y). Then the importance of X to Y is 0 because K(Y|X) = K(Y).

> Superstition is what we use to glue an incomprehensibly complex reality together behind our eyeballs and/or a supercomputer console.

A superstitious compressor C is one that finds statistical patterns in random data and makes incorrect predictions as a result. If X is random then C(X) > |X| = K(X). This is a practical problem in the design of compression algorithms. It gets worse as the model becomes more intelligent and able to recognize more complex patterns.

-- Matt Mahoney, matmahoney@yahoo.com


>
>From: Steve Richfield <steve.richfield@gmail.com>
>To: agi <agi@v2.listbox.com>
>Sent: Mon, March 8, 2010 3:39:04 PM
>Subject: Re: [agi] Data compression book
>
>>Matt,
>
>
>On Mon, Mar 8, 2010 at 5:56 AM, Matt Mahoney <matmahoney@yahoo.com> wrote:
>
>Generalizing the concatenation of XY to a stream of bits, we can prove Steve's point. Important data is compressible, and unimportant data is not.
>
>You're welcome.
>
>And, you have clarified some of my own thoughts.
>
>Point #1:
>
>There is a logical fracture in many of the AGI discussions on this and other forums, to the effect that if a machine understands everything, that it can then better understand speech, can better compress data, etc. Of course, if it understands everything, then it would completely understand the data, and hence would have no need to compress it. Compression deals with the world of PARTIAL understanding, and hence necessarily must stumble over superstitious learning.
>
>So, is (erroneous) superstitious learning a bad thing when it erroneously describes the world? Yes when it impairs compression and prediction, and no when it facilitates compression and prediction.
>
>I have a Christian Science friend with whom I sometimes get into debates on health issues. Our paradigms are as different as two paradigms can get, yet many of our conclusions are the same. Obviously, when viewed from a century in the future, both paradigms will be seen as primitive and crude. Both necessarily carry many errors, but both help "understand" things, though often the "understanding" would be seen as wrong from a future point of view, and by many present observers.
>
>Superstition is what we use to glue an incomprehensibly complex reality together behind our eyeballs and/or a supercomputer console.
>
>Point #2:
>
>This is a simple point, but requires an open mind to "get":
>
>A reasonable initial presumption regarding the operation of our minds, our computers, and our world is that things are as perfect as is conceivable. I have used this simple presumption in metabolic control to solve heretofore unsolvable health problems, by simply presuming that I would have made the same mistakes if I were in someone's metabolic control system, and then looking at what it would take to effectively mislead a VERY bright controller.
>
>This is almost inseparable from a religious belief in God or whatever, because the level of perfection in many/most/all? real world systems so exceeds our own metal facilities that they could be seen as divine, or at least presuming divinity may NOT lead to lots of wrong conclusions or poor compressibility, unless of course you have a crappy belief system.
>
>Relating points 1 and 2 together:
>
>There is probably no direct path from stupid to superintelligent, thereby dooming most present AGI efforts. However, there may just be a path directly to superintelligence by fiat/presumption of divine perfection in our world, and then dealing only with the apparent exceptions. Of course, those "exceptions" are more likely errors in our conception of divinity, rather than "noise" as has been discussed on this thread.
>
>We have been talking about "compression" in terms of dealing with the "noise", i.e. that component that evades our present compression methodologies, when perhaps/probably we should be looking for new rules to better compress/understand the noise. More contorted rules will probably not do well at compression, so there would always be a "gray area" between important "compressible" data and unimportant "incompressible" data, where "noisy" data is still compressed, but poorly.
>
>Hence, importance is measured by compressibility as Matt indicated, but then again this is the SAME measure used to measure LACK of information content, and such input would be seen to carry LOW information content. Lower information content implies greater importance? Perhaps, but there must obviously be some scaling applied, e.g. measuring the "importance" by looking at how much it compresses to? Clearly, more thought is needed here. Matt, can you continue this thought?
>
>Any comments?
>
>Steve
>============
>
>From: Steve Richfield
>> <steve.richfield@gmail.com>
>>
>>To: agi <agi@v2.listbox.com>
>>Sent: Sun, March 7, 2010 4:48:49 PM
>>
>>Subject: Re: [agi] Data compression book
>>
>>
>>>>Regarding Importance:
>>
>>While this may be a "small" factor, the larger overlying issue is how much can a compressed version be shortened by allowing "errors". Here is my feeble attempt to explain the non-obvious part of this discussion:
>>
>>Compression requires understanding of some sort - with better "understanding" leading to better compression. However, when we attempt to compress real-world things like images, we have a combination of the intended plus the unintended/noise. For example, consider compressing Wikipedia. It takes MORE bits to encode the bad grammar, misspellings, ambiguous statements, etc. A "smart" encoder would first produce the "distilled" version that clearly conveys the essence, and then tack on the "noise" needed to turn that into the crap that is now on Wikipedia.
>>
>>To GREATLY shorten the compressed version of Wikipedia, all we would have to do is to strip off the crap at the end. THAT would be a good lossey compression, and might even produce a superior (to the original) result.
>>
>>Note that MPEG actually IMPROVED Starship Troopers, as it eliminated a lot of the dirt and dust from the futuristic scenes, making them look even more perfect.
>>
>>Note the operation of JPEG, where shorter files are obtained by letting JPEG toss the "little" details at its own discression WITHOUT much attention to what is important to us, but rather, what is irksome to the JPEG encoder.
>>
>>Hence, I believe that this discussion has taken a wrong turn. Where it (erroneously) appears that something important may be getting stripped, the REAL problem is that the encoder/decoder lacks deep enough understanding sufficient to utilize this information to produce a better compression, regardless of whether it is lossey or not.
>>
>>If THIS doesn't draw fire, then nothing will.
>>
>>Steve
>>==============
>>
>>On Sat, Mar 6, 2010 at 7:53 AM, Bradley Thomas <brad36@gmail.com> wrote:
>>
>>>What is unimportant? Importance can depend on your viewpoint.
>>>>>>
>>>
>>>
>>>Yes, exactly. See Dessalles' on Simplicity Theory, and that's one of the reasons he posits Simplicity Theory as better than Kolmogorov Complexity. The interestingness of a pattern, i.e. the unexpectedness that arises from a complexity drop, depends on the viewer's prior knowledge. For example, there is the example of a reptile with 200 chromosomes. A layman would think "so what?", but to an expert this is unusual.
>>>
>>>
>>>
>>>On Wed, Mar 3, 2010 at 10:50 AM, Ian Parker <ianparker2@gmail.com> wrote:
>>>
>>>>>>>2. Discard unimportant data.
>>>>
>>>>
>>>>What is unimportant? Importance can depend on your viewpoint. If you are talking about text compression it means all the stylistic information is eliminated. This information is fantastically important for scholarship, establishing authorship, establishing chronologies. Not so different in principle to looking at Evolution with DNA and ignoring the fossil record.
>>>>
>>>>
>>>>One point, the division into important and unimportant? information is in fact a real pons assinorum. If this were done AI/AGI would become a real scholar's assistant.
>>>>
>>>>
>>>>
>>>>
>>>> - Ian Parker
>>>>
>>>>
>>>>
>>>>On 3 March 2010 13:56, Matt Mahoney <matmahoney@yahoo.com> wrote:
>>>>
>>>>John G. Rose wrote:
>>>>>>>>>>> Progressive mode jpeg incrementally reconstitutes an image which has had
>>>>>>>>>>> details thrown away.
>>>>>
>>>>>Well of course the same principle could be applied to lossless compression. Lossy compression has 3 steps.
>>>>>
>>>>>>>>>>1. A transform to separate important and unimportant data.
>>>>>>>>>>2. Discard unimportant data.
>>>>>>>>>>3. Lossless compression of the important data.
>>>>>
>>>>>>>>>>So progressive mode only changes step 3 to order the data from more important to less important, but still important.
>>>>>
>>>>>>>>>>And the reason it applies to lossless is because there really is no such thing as lossless image compression. You are approximating a continuous 2-D field by sampling into pixels and quantizing each sample at discrete levels.
>>>>>
>>>>>
>>>>>>>>>> -- Matt Mahoney, matmahoney@yahoo.com
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>----- Original Message ----
>>>>>>>>>>> From: John G. Rose <johnrose@polyplexic.com>
>>>>>>>>>>> To: agi <agi@v2.listbox.com>
>>>>>
>>>>>> Sent: Tue, March 2, 2010 9:56:18 PM
>>>>>>>>>>> Subject: RE: [agi] Data compression book
>>>>>>>>>>>
>>>>>
>>>>>> > From: Matt Mahoney [mailto:matmahoney@yahoo.com]
>>>>>>>>>>> >
>>>>>
>>>>>> > John G. Rose wrote:
>>>>>>>>>>> > > Is there a strict mathematical definition of compression? Lossy and
>>>>>>>>>>> > lossless
>>>>>>>>>>> > > that you can point to?
>>>>>>>>>>> >
>>>>>>>>>>> > Well, there's my book :)
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Your book does give a general definition of compression.
>>>>>>>>>>>
>>>>>>>>>>> >
>>>>>>>>>>> > > Forget about infinite strings for now, like Pi. Say you have a formula
>>>>>>>>>>> > that
>>>>>>>>>>> > > produces a particular finite string, but it is computationally
>>>>>>>>>>> > infeasible to
>>>>>>>>>>> > > perform the full operation (takes too long). Yet when run it produces
>>>>>>>>>>> > a
>>>>>>>>>>> > > lossless subset of what would be produced if let to run (decompress)to
>>>>>>>>>>> > a
>>>>>>>>>>> > > finished state. It is here that some lossy compressions approach
>>>>>>>>>>> > lossless.
>>>>>>>>>>> > > This is a simple 1 dimensional and basic example of the union of lossy
>>>>>>>>>>> > and
>>>>>>>>>>> > > lossless :)
>>>>>>>>>>> >
>>>>>>>>>>> >
>>>>>>>>>>> > You're describing progressive mode JPEG. The high bits of the low
>>>>>>>>>>> > frequency DCT coefficients are sent first so your browser can start
>>>>>>>>>>> > displaying a blocky approximation of the image before the rest of it is
>>>>>>>>>>> > sent.
>>>>>>>>>>> >
>>>>>>>>>>>
>>>>>>>>>>> Progressive mode jpeg incrementally reconstitutes an image which has had
>>>>>>>>>>> details thrown away. All of the data is not there and is not 100%
>>>>>>>>>>> reproducible, it is only approximated and is catered to the human eye. I'm
>>>>>>>>>>> not sure if there is a progressive mode lossless jpeg but say there was, it
>>>>>>>>>>> would be similar to my above example in the fact that the decompressor would
>>>>>>>>>>> have to give immediate results due to computational constraints and the
>>>>>>>>>>> results would be lossy, resulting from a partial lossless decompression.
>>>>>>>>>>>
>>>>>>>>>>> Many lossy compressions seem to have lossless components, like core
>>>>>>>>>>> information which is saved based on the degree of lossy compression applied.
>>>>>>>>>>> Though this information can be squelched away unless the compressor somehow
>>>>>>>>>>> detects it and marks it non-discardable. A compressor that detects notes in
>>>>>>>>>>> music, or say OCR and speech recognition can be regarded as both lossy and
>>>>>>>>>>> lossless simultaneously, though people may not call them compressors.
>>>>>>>>>>>
>>>>>>>>>>> John
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> -------------------------------------------
>>>>>>>>>>> agi
>>>>>>>>>>> Archives: https://www.listbox.com/member/archive/303/=now
>>>>>>>>>>> RSS Feed: https://www.listbox.com/member/archive/rss/303/
>>>>>>>>>>> Modify Your Subscription:
>>>>>> https://www.listbox.com/member/?&amp;
>>>>>
>>>>>> Powered by Listbox: http://www.listbox.com
>>>>>
>>>>>
>>>>>
>>>>>>>>>>-------------------------------------------
>>>>>>>>>>agi
>>>>>>>>>>Archives: https://www.listbox.com/member/archive/303/=now
>>>>>>>>>>RSS Feed: https://www.listbox.com/member/archive/rss/303/
>>>>>>>>>>Modify Your Subscription: https://www.listbox.com/member/?&amp;
>>>>>>>>>>
>>>>>Powered by Listbox: http://www.listbox.com
>>>>>
>>>>
>>>>>>>>
>>>>agi | Archives >>>> | Modify >>>> Your Subscription
>>>
>>>
>>>--
>>>Brad Thomas
>>>Founder
>>>www.instansa.com
>>>Twitter @bradleymthomas, @instansa
>>>>>>
>>>Linkedin www.bradleythomas.com
>>>
>>>>>>
>>>agi | Archives >>> | Modify >>> Your Subscription
>>
>>>>
>>agi | Archives >> | Modify >> Your Subscription
>>>
>>
>>agi | Archives >> | Modify >> Your Subscription
>
>>
>agi | Archives > | Modify > Your Subscription

Matt,

On Mon, Mar 8, 2010 at 5:56 AM, Matt Mahoney <matmahoney@yahoo.com> wrote:

> Generalizing the concatenation of XY to a stream of bits, we can prove
> Steve's point. Important data is compressible, and unimportant data is not.
>

You're welcome.

And, you have clarified some of my own thoughts.

Point #1:

There is a logical fracture in many of the AGI discussions on this and other
forums, to the effect that if a machine understands everything, that it can
then better understand speech, can better compress data, etc. Of course, if
it understands everything, then it would completely understand the data, and
hence would have no need to compress it. Compression deals with the world of
PARTIAL understanding, and hence necessarily must stumble over superstitious
learning.

So, is (erroneous) superstitious learning a bad thing when it erroneously
describes the world? Yes when it impairs compression and prediction, and no
when it facilitates compression and prediction.

I have a Christian Science friend with whom I sometimes get into debates on
health issues. Our paradigms are as different as two paradigms can get, yet
many of our conclusions are the same. Obviously, when viewed from a century
in the future, both paradigms will be seen as primitive and crude. Both
necessarily carry many errors, but both help "understand" things, though
often the "understanding" would be seen as wrong from a future point of
view, and by many present observers.

Superstition is what we use to glue an incomprehensibly complex reality
together behind our eyeballs and/or a supercomputer console.

Point #2:

This is a simple point, but requires an open mind to "get":

A reasonable initial presumption regarding the operation of our minds, our
computers, and our world is that things are as perfect as is conceivable. I
have used this simple presumption in metabolic control to solve heretofore
unsolvable health problems, by simply presuming that I would have made the
same mistakes if I were in someone's metabolic control system, and then
looking at what it would take to effectively mislead a VERY bright
controller.

This is almost inseparable from a religious belief in God or whatever,
because the level of perfection in many/most/all? real world systems so
exceeds our own metal facilities that they could be seen as divine, or at
least presuming divinity may NOT lead to lots of wrong conclusions or poor
compressibility, unless of course you have a crappy belief system.

Relating points 1 and 2 together:

There is probably no direct path from stupid to superintelligent, thereby
dooming most present AGI efforts. However, there may just be a path directly
to superintelligence by fiat/presumption of divine perfection in our world,
and then dealing only with the apparent exceptions. Of course, those
"exceptions" are more likely errors in our conception of divinity, rather
than "noise" as has been discussed on this thread.

We have been talking about "compression" in terms of dealing with the
"noise", i.e. that component that evades our present compression
methodologies, when perhaps/probably we should be looking for new rules to
better compress/understand the noise. More contorted rules will probably not
do well at compression, so there would always be a "gray area" between
important "compressible" data and unimportant "incompressible" data, where
"noisy" data is still compressed, but poorly.

Hence, importance is measured by compressibility as Matt indicated, but then
again this is the SAME measure used to measure LACK of information content,
and such input would be seen to carry LOW information content. Lower
information content implies greater importance? Perhaps, but there must
obviously be some scaling applied, e.g. measuring the "importance" by
looking at how much it compresses to? Clearly, more thought is needed here.
Matt, can you continue this thought?

Any comments?

Steve
============

*From:* Steve Richfield <steve.richfield@gmail.com>

*To:* agi <agi@v2.listbox.com>
*Sent:* Sun, March 7, 2010 4:48:49 PM

*Subject:* Re: [agi] Data compression book

Regarding Importance:

While this may be a "small" factor, the larger overlying issue is how much
can a compressed version be shortened by allowing "errors". Here is my
feeble attempt to explain the non-obvious part of this discussion:

Compression requires understanding of some sort - with better
"understanding" leading to better compression. However, when we attempt to
compress real-world things like images, we have a combination of the
intended plus the unintended/noise. For example, consider compressing
Wikipedia. It takes MORE bits to encode the bad grammar, misspellings,
ambiguous statements, etc. A "smart" encoder would first produce the
"distilled" version that clearly conveys the essence, and then tack on the
"noise" needed to turn that into the crap that is now on Wikipedia.

To GREATLY shorten the compressed version of Wikipedia, all we would have to
do is to strip off the crap at the end. THAT would be a good lossey
compression, and might even produce a superior (to the original) result.

Note that MPEG actually IMPROVED *Starship Troopers*, as it eliminated a lot
of the dirt and dust from the futuristic scenes, making them look even more
perfect.

Note the operation of JPEG, where shorter files are obtained by letting JPEG
toss the "little" details at its own discression WITHOUT much attention to
what is important to us, but rather, what is irksome to the JPEG encoder.

Hence, I believe that this discussion has taken a wrong turn. Where it
(erroneously) appears that something important may be getting stripped, the
REAL problem is that the encoder/decoder lacks deep enough understanding
sufficient to utilize this information to produce a better compression,
regardless of whether it is lossey or not.

If THIS doesn't draw fire, then nothing will.

Steve
==============
On Sat, Mar 6, 2010 at 7:53 AM, Bradley Thomas <brad36@gmail.com> wrote:

> >What is unimportant? Importance can depend on your viewpoint.
>
> Yes, exactly. See Dessalles' on Simplicity Theory, and that's one of the
> reasons he posits Simplicity Theory as better than Kolmogorov Complexity.
> The interestingness of a pattern, i.e. the unexpectedness that arises from a
> complexity drop, depends on the viewer's prior knowledge. For example, there
> is the example of a reptile with 200 chromosomes. A layman would think "so
> what?", but to an expert this is unusual.
>
> On Wed, Mar 3, 2010 at 10:50 AM, Ian Parker <ianparker2@gmail.com> wrote:
>
>> 2. Discard unimportant data.
>>
>> What is unimportant? Importance can depend on your viewpoint. If you
>> are talking about text compression it means all the stylistic information is
>> eliminated. This information is fantastically important for scholarship,
>> establishing authorship, establishing chronologies. Not so different in
>> principle to looking at Evolution with DNA and ignoring the fossil record.
>>
>> One point, the division into important and *unimportant?* information is
>> in fact a real *pons assinorum*. If this were done AI/AGI would become a
>> real scholar's assistant.
>>
>>
>> - Ian Parker
>>
>>
>> On 3 March 2010 13:56, Matt Mahoney <matmahoney@yahoo.com> wrote:
>>
>>> John G. Rose wrote:
>>> > Progressive mode jpeg incrementally reconstitutes an image which has
>>> had
>>> > details thrown away.
>>>
>>> Well of course the same principle could be applied to lossless
>>> compression. Lossy compression has 3 steps.
>>>
>>> 1. A transform to separate important and unimportant data.
>>> 2. Discard unimportant data.
>>> 3. Lossless compression of the important data.
>>>
>>> So progressive mode only changes step 3 to order the data from more
>>> important to less important, but still important.
>>>
>>> And the reason it applies to lossless is because there really is no such
>>> thing as lossless image compression. You are approximating a continuous 2-D
>>> field by sampling into pixels and quantizing each sample at discrete levels.
>>>
>>> -- Matt Mahoney, matmahoney@yahoo.com
>>>
>>>
>>>
>>> ----- Original Message ----
>>> > From: John G. Rose <johnrose@polyplexic.com>
>>> > To: agi <agi@v2.listbox.com>
>>> > Sent: Tue, March 2, 2010 9:56:18 PM
>>> > Subject: RE: [agi] Data compression book
>>> >
>>> > > From: Matt Mahoney [mailto:matmahoney@yahoo.com]
>>> > >
>>> > > John G. Rose wrote:
>>> > > > Is there a strict mathematical definition of compression? Lossy and
>>> > > lossless
>>> > > > that you can point to?
>>> > >
>>> > > Well, there's my book :)
>>> >
>>> >
>>> > Your book does give a general definition of compression.
>>> >
>>> > >
>>> > > > Forget about infinite strings for now, like Pi. Say you have a
>>> formula
>>> > > that
>>> > > > produces a particular finite string, but it is computationally
>>> > > infeasible to
>>> > > > perform the full operation (takes too long). Yet when run it
>>> produces
>>> > > a
>>> > > > lossless subset of what would be produced if let to run
>>> (decompress)to
>>> > > a
>>> > > > finished state. It is here that some lossy compressions approach
>>> > > lossless.
>>> > > > This is a simple 1 dimensional and basic example of the union of
>>> lossy
>>> > > and
>>> > > > lossless :)
>>> > >
>>> > >
>>> > > You're describing progressive mode JPEG. The high bits of the low
>>> > > frequency DCT coefficients are sent first so your browser can start
>>> > > displaying a blocky approximation of the image before the rest of it
>>> is
>>> > > sent.
>>> > >
>>> >
>>> > Progressive mode jpeg incrementally reconstitutes an image which has
>>> had
>>> > details thrown away. All of the data is not there and is not 100%
>>> > reproducible, it is only approximated and is catered to the human eye.
>>> I'm
>>> > not sure if there is a progressive mode lossless jpeg but say there
>>> was, it
>>> > would be similar to my above example in the fact that the decompressor
>>> would
>>> > have to give immediate results due to computational constraints and the
>>> > results would be lossy, resulting from a partial lossless
>>> decompression.
>>> >
>>> > Many lossy compressions seem to have lossless components, like core
>>> > information which is saved based on the degree of lossy compression
>>> applied.
>>> > Though this information can be squelched away unless the compressor
>>> somehow
>>> > detects it and marks it non-discardable. A compressor that detects
>>> notes in
>>> > music, or say OCR and speech recognition can be regarded as both lossy
>>> and
>>> > lossless simultaneously, though people may not call them compressors.
>>> >
>>> > John
>>> >
>>> >
>>> >
>>> >
>>> >
>>> >
>>> > -------------------------------------------
>>> > agi
>>> > Archives: https://www.listbox.com/member/archive/303/=now
>>> > RSS Feed: https://www.listbox.com/member/archive/rss/303/
>>> > Modify Your Subscription:
>>> > https://www.listbox.com/member/?&amp;
>>> > Powered by Listbox: http://www.listbox.com
>>>
>>>
>>>
>>> -------------------------------------------
>>> agi
>>> Archives: https://www.listbox.com/member/archive/303/=now
>>> RSS Feed: https://www.listbox.com/member/archive/rss/303/
>>> Modify Your Subscription: https://www.listbox.com/member/?&amp;
>>> Powered by Listbox: http://www.listbox.com
>>>
>>
>> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
>> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
>> <http://www.listbox.com&gt;
>>
>
>
>
> --
> Brad Thomas
> Founder
> www.instansa.com
> Twitter @bradleymthomas, @instansa
> Linkedin www.bradleythomas.com
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>

*agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
<https://www.listbox.com/member/archive/rss/303/&gt; |
Modify<https://www.listbox.com/member/?&>Your Subscription
<http://www.listbox.com&gt;

*agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>

Suppose that X is a formula that relates the mass, radius, and luminosity of stars on the main sequence. Suppose Y is a list of stars giving mass, radius, and luminosity. Suppose C is a function returning compressed size. Then if C(Y|X) < C(Y), then I claim that C understands X. For example, if C understands the formula then 2 of the 3 columns in Y are redundant. If C doesn't understand X then it has to store all 3 columns. If C(Y|X) = C(Y) then X is not important to C for the purpose of modeling Y.

-- Matt Mahoney, matmahoney@yahoo.com


>
>From: Ian Parker <ianparker2@gmail.com>
>To: agi <agi@v2.listbox.com>
>Sent: Mon, March 8, 2010 10:45:20 AM
>Subject: Re: [agi] Data compression book
>
>>That still does not yell you the importance of X, Y.
>
>
>Let me take this as an example. Now suppose I am given a list of the characteristics of stars. "Understanding" this example will compress my stellar data. I need to be told the mass and the fact that it is on the main sequence to be able to calculate the luminosity and radius. That does compress the data. In fact understanding this article will not compress it very much. It will though help us to compress (say) the Hipparchos data.
>
>
>My point is this. Often essays of this type contain a limited amount of "hard" information. They are compressible, but not by very much.
>
>
>The passage in fact comes from Damascus. Now Syria is in fact one of the most secular of Arab states. He said in his introduction that he would discuss the evolution of stars but did not do so. If I say that luminosity is mass + proportion of helium I will have made a further generalization. Again compression, but not by that much. A good exercise would be to derive an OWL file by hand. The size of OWL would be the amount of compression.
>
>
>The goal of AGI, or at any rate one of them, must be to construct OWL without any supervision. Most of the information is in fact in terms of style.
>
>
>
>
> - Ian Parker
>
>
>On 8 March 2010 13:56, Matt Mahoney <matmahoney@yahoo.com> wrote:
>
>Regarding "importance", we can give a precise definition. X is important with respect to predicting Y to the extent that K(Y|X) < K(Y). The importance of X in bits is just K(Y|X) - K(Y), where K(.) is Kolmogorov complexity.
>>
>>
>>"Understanding" X is a measure of the number of bits saved in compressing Y, which in general will be less than the uncomputable "importance". It is also measured in bits, and is computable: C(Y|X) - C(Y) where C(Y) is the compressed size of Y, and C(Y|X) is the compressed size of Y after training the model on X. C(Y|X) can be measured as C(XY) - C(X).
>>
>>
>>If X and Y are independent, then X is not important with respect to predicting Y. Nor can X be understood with respect to predicting Y. The lossless
>> compression problem requires understanding X, but determining that X cannot be understood is not computable. The lossy compression problem requires proving that X cannot be understood so that it may be discarded. This problem is also not computable.
>>
>>
>>Generalizing the concatenation of XY to a stream of bits, we can prove Steve's point. Important data is compressible, and unimportant data is not. Thanks for clarifying my thoughts. I will mention this in my book.
>>
>>
>>
>>-- Matt Mahoney, matmahoney@yahoo.com
>>
>>
>>
>>>
>>>From: Steve Richfield
>>> <steve.richfield@gmail.com>
>>>
>>>To: agi <agi@v2.listbox.com>
>>>Sent: Sun, March 7, 2010 4:48:49 PM
>>>
>>>Subject: Re: [agi] Data compression book
>>>
>>>
>>>>>>Regarding Importance:
>>>
>>>While this may be a "small" factor, the larger overlying issue is how much can a compressed version be shortened by allowing "errors". Here is my feeble attempt to explain the non-obvious part of this discussion:
>>>
>>>Compression requires understanding of some sort - with better "understanding" leading to better compression. However, when we attempt to compress real-world things like images, we have a combination of the intended plus the unintended/noise. For example, consider compressing Wikipedia. It takes MORE bits to encode the bad grammar, misspellings, ambiguous statements, etc. A "smart" encoder would first produce the "distilled" version that clearly conveys the essence, and then tack on the "noise" needed to turn that into the crap that is now on Wikipedia.
>>>
>>>To GREATLY shorten the compressed version of Wikipedia, all we would have to do is to strip off the crap at the end. THAT would be a good lossey compression, and might even produce a superior (to the original) result.
>>>
>>>Note that MPEG actually IMPROVED Starship Troopers, as it eliminated a lot of the dirt and dust from the futuristic scenes, making them look even more perfect.
>>>
>>>Note the operation of JPEG, where shorter files are obtained by letting JPEG toss the "little" details at its own discression WITHOUT much attention to what is important to us, but rather, what is irksome to the JPEG encoder.
>>>
>>>Hence, I believe that this discussion has taken a wrong turn. Where it (erroneously) appears that something important may be getting stripped, the REAL problem is that the encoder/decoder lacks deep enough understanding sufficient to utilize this information to produce a better compression, regardless of whether it is lossey or not.
>>>
>>>If THIS doesn't draw fire, then nothing will.
>>>
>>>Steve
>>>==============
>>>
>>>On Sat, Mar 6, 2010 at 7:53 AM, Bradley Thomas <brad36@gmail.com> wrote:
>>>
>>>>What is unimportant? Importance can depend on your viewpoint.
>>>>>>>>
>>>>
>>>>
>>>>Yes, exactly. See Dessalles' on Simplicity Theory, and that's one of the reasons he posits Simplicity Theory as better than Kolmogorov Complexity. The interestingness of a pattern, i.e. the unexpectedness that arises from a complexity drop, depends on the viewer's prior knowledge. For example, there is the example of a reptile with 200 chromosomes. A layman would think "so what?", but to an expert this is unusual.
>>>>
>>>>
>>>>
>>>>On Wed, Mar 3, 2010 at 10:50 AM, Ian Parker <ianparker2@gmail.com> wrote:
>>>>
>>>>>>>>>2. Discard unimportant data.
>>>>>
>>>>>
>>>>>What is unimportant? Importance can depend on your viewpoint. If you are talking about text compression it means all the stylistic information is eliminated. This information is fantastically important for scholarship, establishing authorship, establishing chronologies. Not so different in principle to looking at Evolution with DNA and ignoring the fossil record.
>>>>>
>>>>>
>>>>>One point, the division into important and unimportant? information is in fact a real pons assinorum. If this were done AI/AGI would become a real scholar's assistant.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> - Ian Parker
>>>>>
>>>>>
>>>>>
>>>>>On 3 March 2010 13:56, Matt Mahoney <matmahoney@yahoo.com> wrote:
>>>>>
>>>>>John G. Rose wrote:
>>>>>>>>>>>>> Progressive mode jpeg incrementally reconstitutes an image which has had
>>>>>>>>>>>>> details thrown away.
>>>>>>
>>>>>>Well of course the same principle could be applied to lossless compression. Lossy compression has 3 steps.
>>>>>>
>>>>>>>>>>>>1. A transform to separate important and unimportant data.
>>>>>>>>>>>>2. Discard unimportant data.
>>>>>>>>>>>>3. Lossless compression of the important data.
>>>>>>
>>>>>>>>>>>>So progressive mode only changes step 3 to order the data from more important to less important, but still important.
>>>>>>
>>>>>>>>>>>>And the reason it applies to lossless is because there really is no such thing as lossless image compression. You are approximating a continuous 2-D field by sampling into pixels and quantizing each sample at discrete levels.
>>>>>>
>>>>>>
>>>>>>>>>>>> -- Matt Mahoney, matmahoney@yahoo.com
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>----- Original Message ----
>>>>>>>>>>>>> From: John G. Rose <johnrose@polyplexic.com>
>>>>>>>>>>>>> To: agi <agi@v2.listbox.com>
>>>>>>
>>>>>>> Sent: Tue, March 2, 2010 9:56:18 PM
>>>>>>>>>>>>> Subject: RE: [agi] Data compression book
>>>>>>>>>>>>>
>>>>>>
>>>>>>> > From: Matt Mahoney [mailto:matmahoney@yahoo.com]
>>>>>>>>>>>>> >
>>>>>>
>>>>>>> > John G. Rose wrote:
>>>>>>>>>>>>> > > Is there a strict mathematical definition of compression? Lossy and
>>>>>>>>>>>>> > lossless
>>>>>>>>>>>>> > > that you can point to?
>>>>>>>>>>>>> >
>>>>>>>>>>>>> > Well, there's my book :)
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> Your book does give a general definition of compression.
>>>>>>>>>>>>>
>>>>>>>>>>>>> >
>>>>>>>>>>>>> > > Forget about infinite strings for now, like Pi. Say you have a formula
>>>>>>>>>>>>> > that
>>>>>>>>>>>>> > > produces a particular finite string, but it is computationally
>>>>>>>>>>>>> > infeasible to
>>>>>>>>>>>>> > > perform the full operation (takes too long). Yet when run it produces
>>>>>>>>>>>>> > a
>>>>>>>>>>>>> > > lossless subset of what would be produced if let to run (decompress)to
>>>>>>>>>>>>> > a
>>>>>>>>>>>>> > > finished state. It is here that some lossy compressions approach
>>>>>>>>>>>>> > lossless.
>>>>>>>>>>>>> > > This is a simple 1 dimensional and basic example of the union of lossy
>>>>>>>>>>>>> > and
>>>>>>>>>>>>> > > lossless :)
>>>>>>>>>>>>> >
>>>>>>>>>>>>> >
>>>>>>>>>>>>> > You're describing progressive mode JPEG. The high bits of the low
>>>>>>>>>>>>> > frequency DCT coefficients are sent first so your browser can start
>>>>>>>>>>>>> > displaying a blocky approximation of the image before the rest of it is
>>>>>>>>>>>>> > sent.
>>>>>>>>>>>>> >
>>>>>>>>>>>>>
>>>>>>>>>>>>> Progressive mode jpeg incrementally reconstitutes an image which has had
>>>>>>>>>>>>> details thrown away. All of the data is not there and is not 100%
>>>>>>>>>>>>> reproducible, it is only approximated and is catered to the human eye. I'm
>>>>>>>>>>>>> not sure if there is a progressive mode lossless jpeg but say there was, it
>>>>>>>>>>>>> would be similar to my above example in the fact that the decompressor would
>>>>>>>>>>>>> have to give immediate results due to computational constraints and the
>>>>>>>>>>>>> results would be lossy, resulting from a partial lossless decompression.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Many lossy compressions seem to have lossless components, like core
>>>>>>>>>>>>> information which is saved based on the degree of lossy compression applied.
>>>>>>>>>>>>> Though this information can be squelched away unless the compressor somehow
>>>>>>>>>>>>> detects it and marks it non-discardable. A compressor that detects notes in
>>>>>>>>>>>>> music, or say OCR and speech recognition can be regarded as both lossy and
>>>>>>>>>>>>> lossless simultaneously, though people may not call them compressors.
>>>>>>>>>>>>>
>>>>>>>>>>>>> John
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> -------------------------------------------
>>>>>>>>>>>>> agi
>>>>>>>>>>>>> Archives: https://www.listbox.com/member/archive/303/=now
>>>>>>>>>>>>> RSS Feed: https://www.listbox.com/member/archive/rss/303/
>>>>>>>>>>>>> Modify Your Subscription:
>>>>>>> https://www.listbox.com/member/?&amp;
>>>>>>
>>>>>>> Powered by Listbox: http://www.listbox.com
>>>>>>
>>>>>>
>>>>>>
>>>>>>>>>>>>-------------------------------------------
>>>>>>>>>>>>agi
>>>>>>>>>>>>Archives: https://www.listbox.com/member/archive/303/=now
>>>>>>>>>>>>RSS Feed: https://www.listbox.com/member/archive/rss/303/
>>>>>>>>>>>>Modify Your Subscription: https://www.listbox.com/member/?&amp;
>>>>>>>>>>>>
>>>>>>Powered by Listbox: http://www.listbox.com
>>>>>>
>>>>>
>>>>>>>>>>
>>>>>agi | Archives >>>>> | Modify >>>>> Your Subscription
>>>>
>>>>
>>>>--
>>>>Brad Thomas
>>>>Founder
>>>>www.instansa.com
>>>>Twitter @bradleymthomas, @instansa
>>>>>>>>
>>>>Linkedin www.bradleythomas.com
>>>>
>>>>>>>>
>>>>agi | Archives >>>> | Modify >>>> Your Subscription
>>>
>>>>>>
>>>agi | Archives >>> | Modify >>> Your Subscription
>>>>
>>agi | Archives >> | Modify >> Your Subscription
>
>>
>agi | Archives > | Modify > Your Subscription

Dorian, what is your academic affiliation at the moment? I found on the net
that you were a postodc at Stanford. Where are you now? And who else works
on NED in the University system?
Thanks,
Giovanni


On Tue, Mar 2, 2010 at 3:32 PM, Dorian Aur <dorianaur@gmail.com> wrote:

> Dear Colleagues,
>
>
>
> We are very happy to announce the publication of our book:
>
> *Neuroelectrodynamics: Understanding the Brain Language (2010)*
>
> By Dorian Aur and Mandar S.Jog, IOS Press, 2010
>
>
>
> *Description from the publisher: *
>
>
>
>
> The essence of brain function consists in how information is processed,
>
> transferred and stored. Current neurophysiological doctrine remains focused
>
> within a spike timing paradigm, but this has a limited capacity for
> advancing
>
> the understanding of how the brain works. This book puts forward a new
>
> model; the neuroelectrodynamic model (NED), which describes the intrinsic
>
> computational processes by the dynamics and interaction of charges. It uses
>
> established laws of physics, such as those of classical mechanics,
>
> thermodynamics and quantum physics, as the guiding principle to develop a
>
> general theoretical construct of the brain’s computational model, which
>
> incorporates the neurobiology of the cells and the molecular machinery
> itself,
>
> along with the electrical activity in neurons, to explain experimental
> results
>
> and predict the organization of the system. After addressing the
> deficiencies
>
> of current approaches, the laws and principles required to build a new
> model
>
> are discussed.
>
> In addition, as well as describing experiments which provide the required
> link
>
> between computation and semantics, the book highlights important concepts
>
> relating the theory of information with computation and the electrical
>
> properties of neurons. The NED model is explained and expounded and
>
> several examples of its application are shown. Of interest to all those
>
> involved in the fields of neuroscience, neurophysiology, computer science
>
> and the development of artificial intelligence, NED is a step forward in
>
> understanding the mind in computational terms.
>
> Please, visit the book’s website for more information and pricing (
> http://www.iospress.nl/loadtop/load.php?isbn=9781607500919)
>
> or (http://www.iospress.nl/flyers_b/fl9781607500919.pdf)
>
>
>
> *CHAPTER 1. Understanding the Brain Language*
>
> • Myths, Controversies and Current Challenges
>
> • Spike Timing – Current Challenges
>
> • Issues of Brain Computation
>
> • How can Information be Represented in the Brain
>
> • Interdisciplinary Formalism
>
> *CHAPTER 2. Imaging Spikes – Challenges for Brain Computations*
>
> • Methods, Models and Techniques
>
> • From Spikes to Behavior
>
> *CHAPTER 3. Information and Computation*
>
> • What is Computation?
>
> • Neurons as information engines
>
> *CHAPTER 4. Models of Brain Computation*
>
> • Dynamics Based Computation – The Power of Charges
>
> • Quantum Model – Combining Many Worlds
>
> • The Thermodynamic Model of Computation
>
> *CHAPTER 5. From Brain Language to Artificial Intelligence*
>
> • How are memories stored?
>
> • Spike Timing – An Incomplete Description
>
> • Instead of Discussion
>
>
>
> Comments and critiques are most welcome!
> ** Apologies for cross-posting **
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>

Is there a discount for the member of this list? ; )

Giovanni

On Tue, Mar 2, 2010 at 3:32 PM, Dorian Aur <dorianaur@gmail.com> wrote:

> Dear Colleagues,
>
>
>
> We are very happy to announce the publication of our book:
>
> *Neuroelectrodynamics: Understanding the Brain Language (2010)*
>
> By Dorian Aur and Mandar S.Jog, IOS Press, 2010
>
>
>
> *Description from the publisher: *
>
>
>
>
> The essence of brain function consists in how information is processed,
>
> transferred and stored. Current neurophysiological doctrine remains focused
>
> within a spike timing paradigm, but this has a limited capacity for
> advancing
>
> the understanding of how the brain works. This book puts forward a new
>
> model; the neuroelectrodynamic model (NED), which describes the intrinsic
>
> computational processes by the dynamics and interaction of charges. It uses
>
> established laws of physics, such as those of classical mechanics,
>
> thermodynamics and quantum physics, as the guiding principle to develop a
>
> general theoretical construct of the brain’s computational model, which
>
> incorporates the neurobiology of the cells and the molecular machinery
> itself,
>
> along with the electrical activity in neurons, to explain experimental
> results
>
> and predict the organization of the system. After addressing the
> deficiencies
>
> of current approaches, the laws and principles required to build a new
> model
>
> are discussed.
>
> In addition, as well as describing experiments which provide the required
> link
>
> between computation and semantics, the book highlights important concepts
>
> relating the theory of information with computation and the electrical
>
> properties of neurons. The NED model is explained and expounded and
>
> several examples of its application are shown. Of interest to all those
>
> involved in the fields of neuroscience, neurophysiology, computer science
>
> and the development of artificial intelligence, NED is a step forward in
>
> understanding the mind in computational terms.
>
> Please, visit the book’s website for more information and pricing (
> http://www.iospress.nl/loadtop/load.php?isbn=9781607500919)
>
> or (http://www.iospress.nl/flyers_b/fl9781607500919.pdf)
>
>
>
> *CHAPTER 1. Understanding the Brain Language*
>
> • Myths, Controversies and Current Challenges
>
> • Spike Timing – Current Challenges
>
> • Issues of Brain Computation
>
> • How can Information be Represented in the Brain
>
> • Interdisciplinary Formalism
>
> *CHAPTER 2. Imaging Spikes – Challenges for Brain Computations*
>
> • Methods, Models and Techniques
>
> • From Spikes to Behavior
>
> *CHAPTER 3. Information and Computation*
>
> • What is Computation?
>
> • Neurons as information engines
>
> *CHAPTER 4. Models of Brain Computation*
>
> • Dynamics Based Computation – The Power of Charges
>
> • Quantum Model – Combining Many Worlds
>
> • The Thermodynamic Model of Computation
>
> *CHAPTER 5. From Brain Language to Artificial Intelligence*
>
> • How are memories stored?
>
> • Spike Timing – An Incomplete Description
>
> • Instead of Discussion
>
>
>
> Comments and critiques are most welcome!
> ** Apologies for cross-posting **
> *agi* | Archives <https://www.listbox.com/member/archive/303/=now&gt;
> <https://www.listbox.com/member/archive/rss/303/&gt; | Modify<https://www.listbox.com/member/?&>Your Subscription
> <http://www.listbox.com&gt;
>

If you'd like more contributions, why not post some of your brainstorming on
here? - The more the channels of this group deversify, the more difficult
it gets to follow - and the more likely it gets to break up altogether.

--------------------------------------------------
From: "YKY (Yan King Yin, ???)" <generic.intelligence@gmail.com>
Sent: Monday, March 08, 2010 5:38 AM
To: "agi" <agi@v2.listbox.com>
Subject: Re: [agi] brainstorming contests?

> On Mon, Mar 8, 2010 at 1:06 PM, The Wizard <key.universe@gmail.com> wrote:
>>
>> What are you guys brainstorming about?
>
> Can be anything related to AGI, for example today I talked about a new
> approach to AGI where we develop a stochastic search algorithm (such
> as evolutionary search) and use it to search for the AGI program, and
> we don't directly code the AGI, but instead use exclusively a new
> "nudging" technique to steer the code to desired states,
> interactively. The idea may not be practical, but it's fun.
>
> YKY
>
>
> -------------------------------------------
> agi
> Archives: https://www.listbox.com/member/archive/303/=now
> RSS Feed: https://www.listbox.com/member/archive/rss/303/
> Modify Your Subscription:
> https://www.listbox.com/member/?&amp;
> Powered by Listbox: http://www.listbox.com