The last invention.

  • 388 Replies
  • 50978 Views
*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *************
  • Transformer
  • *
  • 2108
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #270 on: February 02, 2018, 01:10:05 am »
wait....if each column is 1 detector....then why 6 layers? 6 layers implies line-to-face hierarchy...

If all of a column stands for 1 representation/feature/concept, then there is no layers in the neo haha oh come on...I mean you can emulate the layer thing by connecting the columns but.....so...right? And isn't that stupid?
« Last Edit: February 02, 2018, 09:00:10 am by LOCKSUIT »
Emergent

*

Bob

  • Roomba
  • *
  • 3
Re: The last invention.
« Reply #271 on: February 02, 2018, 12:12:37 pm »
Well, most input from the thalamus enters the columns in layer 4 - about in the middle. So a line-to-face hierarchy in a single column doesn't make much sense to me.

Why do you think it is stupid? Or more stupid than having a line-to-face hierarchy in a single column (there would be some serious dense-packed magic processing if everything happened in such a small space which we can't make sense of). Now given that a column detects only one feature, that doesn't mean the different layers aren't there for a reason. They process/do something with the information in some way. I guess you could emulate the layer thing. But what do you connect to what? And you still need to implement what happens inside the columns.

But I would love to hear korrelan on this and on how he has implemented it.

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *************
  • Transformer
  • *
  • 2108
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #272 on: February 02, 2018, 09:45:03 pm »
see attachment

This expresses my feeling and "image" I'm experiencing.
« Last Edit: February 02, 2018, 10:28:19 pm by LOCKSUIT »
Emergent

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1083
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #273 on: February 03, 2018, 05:47:30 pm »
Most current estimates state that there are approximately 86 billion neurons total in the human brain with around 16 billion neurons in the cerebral cortex.  Some animals have more neurons total but we have the highest quota in the cerebral cortex leading academics to presume this is the seat of our intelligence.

According to the current level of understanding by academia, the human cortex seems to have at most six layers, it depends on how you define/ perceive the laminar structure; some areas have less than six.  The cortex is comprised of modular units, a similar repeating pattern that gives the appearance of functional modules/ columns.  There are approximately two million cortical/ hyper columns, each column is comprised of approximately 80 mini columns, and each mini column has 80 to 110 neurons distributed through the six (or less) layers.

Ok, there is some good news… and some bad news lol.

The good news is that cortical columns do exist in the human cortex; they do have a modular function and act as distinct processing units.  As we get older our cortex thins and the columns get further apart… and this gives you a clue as to the bad news.

The bad news is that you can’t use the columnar structure as a guide to figuring out how the brain learns because they are not an innate structure; they are not produced by ‘evolution/ DNA’… but by experience/ learning.  The columns don’t adhere to a standard recognisable format across an average population of cortex's; everyone’s columnar organisation is going to be totally different.  The columns are a product of self organisation/ learning/ experience.

Of course this gives us a target/ clue as to the wiring/ coding/ transmission protocols the brain uses, just because the correct schema should produce columns out of a regular laminate sheet of neurons with random selective fields.

I produced this short vid last night showing the formation of columns in a random sheet of neurons using my AGI’s internal wiring schema. The sheet is subjected to 80 GTP pattern facets, as it self organises from the data stream you can see the columns forming, represented by the colour boundaries.



 :)
« Last Edit: February 04, 2018, 09:02:52 am by korrelan »
It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *************
  • Transformer
  • *
  • 2108
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #274 on: February 03, 2018, 10:33:38 pm »
Korrelan you gotta answer my question though, you must have missed it above.

See my drawing above? Which way is the human brain? If not the right side (like ANNs are setup), then it's gotta be like the left side of the drawing, so that it is a hierarchy of higher layers get it. However, if it is the left side of my drawing then, that makes me sad cus, it not cool....
Emergent

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1083
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #275 on: February 04, 2018, 10:13:05 am »
Hi Lock

Actually neither is correct but the closest I’m afraid is the left diagram.  Layer one is at the top by the way, with layer six being the deepest.

You seem to have mixed up the function of the six cortical layers with the layers in a CNN pooling network, its not one layer per recognition process.

As Bob described, the six layers are used as processing mechanism, not as recognition layers.  The layers are used by the structure/ connectome of the brain to process/ blend information from many different other cortical regions and deep brain structures.

The columns represent very abstract facets of recognition, even in the visual cortex its not just line/ gradient orientation columns, the columns change their function depending on the overall state of the GTP/ current thought pattern.  Even the state of the surrounding columns can change the functionality of a single column.

keep in mind that everyone has different opinions and theories on how the brain functions, even academia can't agree on most aspects of the layout and functions of the brains regions.  Anything I state is just my opinion from my research/ reading and experimentation.  There is no reason why you couldn't use the layout in your diagrams to create a theory, if you can figure away to make it all work.  I choose to stick as closely as possible to our own brains design, this is not necessarily required to build a functioning AGI.

 :)
« Last Edit: February 04, 2018, 10:41:49 am by korrelan »
It thunk... therefore it is!

*

ivan.moony

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1108
    • Some of my projects
Re: The last invention.
« Reply #276 on: February 04, 2018, 10:41:00 am »
I imagine NN as a kind of blackbox for detecting features. How you connect features to hierarchy like (face) / (hair - eyes - nose - mouth - ears) and so on deeper, is about natural programming being done outside the blackbox. Now, this outside programming can also be done by outer NN, recursively. At the end, we would have |NN inside NN inside NN ...| for |features inside features inside features ...|

If I am right, than it is possible to recursively structure NNs to any level depth, producing the left Locksuit's image effect.

But maybe I'm wrong...
Dream big. The bigger the dream is, the more beautiful place the world becomes.

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1083
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #277 on: February 04, 2018, 10:50:33 am »
@Ivan

I agree, part of the brains recognition process is a parallel recursive pattern.  But the pattern does not move down through the layers of the cortex, with each layer selecting a different facet, it moves through the whole structure of the cortex and lower deep structures.

The ‘deepness’ of the recognition cycle does not come from the deepness of the layer but from the time the cortex has had to process/ recognise the pattern… if that makes sense.

 :)
It thunk... therefore it is!

*

ivan.moony

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1108
    • Some of my projects
Re: The last invention.
« Reply #278 on: February 04, 2018, 10:59:23 am »
Is it the case that:
  • the same NN trained data is used for recognizing any of a number of features, no matter of how the features form a hierarchy
  • multiple layers are used only for adjusting the accuracy of a single NN (sometimes less number of layers give better results), without forming a hierarchy between recognized objects
Dream big. The bigger the dream is, the more beautiful place the world becomes.

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *************
  • Transformer
  • *
  • 2108
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #279 on: February 04, 2018, 08:39:04 pm »
I put it together why korrelan didn't really reply back to me yesterday LOL, he was drinking again lol! As he stated so.

If a neuron dies, the others right beside it in the same column do the same job - they ALL stand for ex. a line or nose or face = redundancy/robustness/accuracy/reliableness. Next, it doesn't really matter what order they are in i.e. higher levels, the connections are the SAME.

But I seriously want to know why Wiki / Ray Kurzweil said "and it goes UP the layers to higher concepts".

Also, so the input to the columns starts at the top? Or layer 4 on the SIDE of the column?

Wait a minute....Also, Wiki shows ex. V1 being for more simple features at the back of the human brain, and by the time it gets to IT near the side of the temporal area it is highhhher feature representations! So they are grouped! But sideways! Right?

See attachment below:
« Last Edit: February 05, 2018, 03:55:55 am by LOCKSUIT »
Emergent

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1083
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #280 on: February 05, 2018, 11:10:39 am »
@Ivan

Quote
the same NN trained data is used for recognizing any of a number of features, no matter of how the features form a hierarchy

If someone was to take a pencil and draw/ outline a letter of the alphabet on your face cheek you would be able to recognise the letter straight away, with no prior experience. You can do the same easily for any area of your skins surface.  This gives us quite a few clues to the processes involved in the overall recognition schema.  Past the initial sensing on the skins surface the information is being converted into a commonly understood pattern that represents a character of the alphabet.  You only have a few million cortical columns at most; there just are not enough neurons in your brain for each area of skin to have its own recognition hierarchy.

Quote
multiple layers are used only for adjusting the accuracy of a single NN (sometimes less number of layers give better results), without forming a hierarchy between recognized objects

Yes, the layers are part of the mechanism that allows the cortical regions to recognise stimulus.  If you insert a probe down though the layers perpendicular to the surface then all the neurons in that column\ stack have the same basic receptive field layout.  Think of each column as a cog in the machine.  You global thought pattern GTP is constantly cycling though your connectome, the sensory cortex regions take in external stimulus, recognise the stimulus, convert it into a sparse pattern that depends on the current state/ feedback within the GTP and injects the pattern back into the GTP.  This affects the GTP which in turn changes your perception/ thought train, which then affects the recognition of the next sensory frame… repeat.

@Lock

Yes I was having a drink lol.  :D

Mr Kurzweil is talking about layers within the hierarchal learning structure/ schema, not the layers within the cortex.

Most sensory stimulus does indeed enter the cortex at layer four/ five, though this is not through the side lol.  Columns are never cylindrical, and there is no actual physical space between them.  The boundaries between columns are mostly filled with inhibitory neurons as well as the white matter lateral connections, glial cells, etc.  They are sometimes drawn as cylindrical just to aid understanding.  The afferent axons that carry sensory stimulus into layer four pass through layers 6, 5 and indeed sometimes make synapse to these layers.  The cortex is a continuous sheet, physically sub divided only by the lobe boundaries and a column is just a small area that has specialised in a particular trait, it’s not an actual physical structure… it’s a logical/ functional structure.

Quote
So they are grouped! But sideways! Right?

Yes.  In certain areas of the cortex sensory stimulus can be mapped as moving across the surface to adjacent areas, and research seems to link this with the hierarchy of recognition. Though keep in mind that other deep brain structures are also playing their part, as is the rest of the cortex.

 :)
It thunk... therefore it is!

*

johnphantom

  • Roomba
  • *
  • 9
Re: The last invention.
« Reply #281 on: February 06, 2018, 02:04:40 pm »
"The beauty is that the emergent connectome defines both the structural hardware and the software.  The brain is more like a clockwork watch or a Babbage engine than a modern computer.  The design of a cog defines its functionality.  Data is not passed around within a watch, there is no software; but complex calculations are still achieved.  Each module does a specific job, and only when working as a whole can the full and correct function be realised. (Clockwork Intelligence: Korrelan 1998)"

I have created a working model for a stateless computer. It is pure connectionism, which I look at as a geometry of information. Using quantum nonlocality it could operate instantaneously as input happens:

http://tinyurl.com/statelesscomputer

or directly:

https://app.box.com/...p8tir00r0pf1467

Note that this site does not support advertising and there is nothing that you have to download to read what I wrote.

You can contact me at johnphantom@hotmail.com

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1083
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #282 on: February 15, 2018, 03:56:23 pm »



The ideal shape/ volume I’ve found so far for the cortex connectome model is spherical but the four lobes per hemisphere, the separate hemispheres themselves and even the gyri/ sulci are still required, they all have a functional purpose within the model.

This is an experiment to roughly map my models connectome to the same area/ shape as the human cortex.  Besides the procrastination element to this experiment, the idea is that it should aid understanding of the schema to the layman… and I think it looks cool lol.

Each voxel = one functional column (50 ish neurons, 1000 ish synapse) per 10.5K voxels.

Although even this fetal stage can learn hundreds of millions patterns I’m sticking to the usual 40 distinct colours for clarity.  40 patterns learnt * 250 facets = 10k pattern facets learned.

As usual the voxel colours match the bar graph lower left.  Each colour represents one injected/ learned full pattern.  As the patterns are injected into the model (top right) the height of the bars represent confidence in recognising that pattern.  Ideally just one bar should rise for each pattern. If more than one bar rises then a similarity in the patterns exists.

 :)
« Last Edit: February 16, 2018, 10:56:33 am by korrelan »
It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *************
  • Transformer
  • *
  • 2108
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #283 on: February 17, 2018, 11:51:20 pm »
Omg I'm so excited now...Did you just make your project into the shape of the human brain!? For a project like this that's trying to re-create the human mind/brain, that is definitely a step forward.
Emergent

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *************
  • Transformer
  • *
  • 2108
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #284 on: March 14, 2018, 11:04:23 pm »
Oh so GTP is C, A, T being heard? The selection keeps updates each step.

And after hearing "CAT", it echoes, and this updates the GTP each step, which also updates the GTP each step!! WHICH ALSO... ? What's the point in that though? It's the same word. I know hearing CAT 6 times is a new "representation" or "word" or "sentence" but it has no meaning/use in English Vocabulary...unless, I was never told......

CAT
CAT CAT CAT CAT
CAT
CAT CAT

See, I don't understand that sentence. 4 cats heard as a GTP self feed-in was useless in this understanding of it.
Emergent

 


FAQ on the newly established MIT Stephen A. Schwarzman College of Computing
by Tyler (Robotics News)
October 17, 2018, 12:00:10 pm
MIT reshapes itself to shape the future
by Tyler (Robotics News)
October 16, 2018, 12:01:15 pm
Tag... you're it...
by Art (AI News )
October 16, 2018, 02:55:01 am
In need of a psychological coach?
by ruebot (AI News )
October 13, 2018, 01:33:37 pm
Leela Chess Zero
by Art (AI News )
October 08, 2018, 03:30:42 pm
Alpha GO Zero
by Art (AI News )
October 08, 2018, 02:48:32 pm
Detecting fake news at its source
by Tyler (Robotics News)
October 06, 2018, 12:00:55 pm
Phase Change Memory
by Art (AI News )
October 05, 2018, 01:21:59 pm

Users Online

75 Guests, 1 User
Users active in past 15 minutes:
ivan.moony
[Trusty Member]

Most Online Today: 163. Most Online Ever: 208 (August 27, 2008, 09:36:30 am)

Articles