The last invention.

  • 317 Replies
  • 34964 Views
*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #300 on: March 20, 2018, 02:02:57 pm »
Now korrelan.

You said your AGI makes discoveries by doing Pattern Recognition between 2 things. One thing is fact. The other thing is new hence 'discovery'. In your AGI, where does the 'new' thing come from to be compared for commonalities to the other thing/fact?

:)
Emergent

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 855
Re: The last invention.
« Reply #301 on: March 20, 2018, 06:08:45 pm »
 Two thing here. The target and the starting object. With simulated annealing you mess with the weight until you start moving to
the target in the most direct way. The focus pointer movement to the target will slowly inch toward the target.
 On the way over the focus recreated with a generative NN or algorithm a graphics gaming engine.
The re creation may will look like a morph from on to the other. But somewhere in between might be a novel discovery.
 Once The straight movement is found, that is correct ratio of incrementing of certain weight, a vector list. Then we
can shoot past the target and find novel things beyond the space in between.
 A second way is to lower the resolutions by lesson in the focus and thing get blurry. Different object will start to look the same.
 In this fashion substitutions of objects in temporal pattern loop can be done. This is a way of discovering new patterns loops.


*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #302 on: March 20, 2018, 06:17:44 pm »
Sounds similar to the answer keghn very good. I think I understand what you're saying.

I'd like to hear korrelan's answer too.
Emergent

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #303 on: March 30, 2018, 02:14:49 am »
Questions for korrelan:

Do you use a RNN? Do you use a LSTM? If not then what?

You said your AGI makes discoveries by doing Pattern Recognition between 2 things. One thing is fact. The other thing is new hence 'discovery'. In your AGI, where does the 'new' thing come from to be compared for commonalities to the other thing/fact?
Emergent

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 905
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #304 on: March 30, 2018, 11:08:49 am »
Apologies I didn’t see the above questions.

Quote
Do you use a RNN? Do you use a LSTM? If not then what?

The simple answer is all of the above.  Each type of common artificial neural network schema has been loosely modelled on the brain utilizing neurons, connections, weights, etc.  Each has its own benefits; each excels at a specific type of recognition.  It was obvious that all these artificial schemas are so similar yet achieve very different tasks, so there must be a way of combining them.

I spent years studying the biological connectome, drawing conclusions, writing the code to simulate the same functions, whilst referencing other people’s pervious research on artificial/ biological networks.  I eventually narrowed it down to a specific connectome/ set of algorithms/ design that can simulate/ achieve all the properties of common artificial schemas… plus a lot more.

So… I basically just took the best bits from each schema, mixed them all together and created a hybrid that has all the properties that I figured a biological neuron must have to achieve what it does.

The connectome constantly evolves/ changes overt time, +/- synapse are generated, culled depending on what the system is experiencing.  The type and properties of neurons/ synapse change according to their age/ location.  A set of algorithms governs the growth and 3D properties of the connectome. It also requires rest and sleep… very important for the schema.

Quote
You said your AGI makes discoveries by doing Pattern Recognition between 2 things. One thing is fact. The other thing is new hence 'discovery'. In your AGI, where does the 'new' thing come from to be compared for commonalities to the other thing/fact?

Good question. 

Firstly it’s important to note that the AGI is a closed system, so it can only use its own cameras, microphones, etc to experience the world.  If you where text chatting with the AGI it would be reading your text off a monitor with its own eyes/ cameras and typing on a keyboard.  A skype call will be a much better way of experiencing the AGI. 

There are many reasons for the closed system, but mainly it negates a lot of complexity that has been incorrectly assumed by AI research.  Humans don’t scan a huge high resolution image for faces/ objects for example, 90% of our ability to recognise a face is centred on the fovea. We move our eyes to the target, our peripheral vision guides the fovea, it is doing a very different job and using different parts of the visual data stream.

So any incoming sensory stream is always modulated from the machines perspective/ point of view.  The sensory signals from the cameras are fixed and unique to its visual systems.

When the AGI recognises a GTP/ concept the facets/ commonalities it’s recognising and combining are many levels of abstraction lower than you would imagine.  If it’s looking at a face everything it can see, the position/ spacing of the eyes, the shading/ gradients, angle of the face, motion, everything is learned/ recognised/ combined at a very low level of abstraction. 

Eventually you end up with a generic face recogniser and the sub facets will generate a GTP pattern for that type of face.  This generic pattern is altered/ enhanced by the unique properties of the face.  Within a few milliseconds you have a GTP pattern built from both common facial properties and the unique properties that make the face stand out.

It is at this level of abstraction that all concepts are initially recognised, the resulting GTP patterns for all levels of resolution/ recognition can be fired from any part of the hierarchical GTP structure the concept comprises.  Give it two red objects and they will both fire the same GTP patterns for ‘red’ and ‘object’ it will recognise/ link the commonality.

 :)
It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #305 on: March 30, 2018, 08:20:01 pm »
Wait so, you mean to tell me that you combined LSTMs, GANs, RBMs, HFNs, etc + realistic human based NN schema... into a hybrid NN?


Oh also, if is the case, please list all you combined in short-form ex. GAN (don't write out the full name). Ohh oh and also give a very short brief reason why you combined each ex:

GAN - generative abilities
LSTM - recurrent abilities
etc...
Emergent

*

infurl

  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 479
  • Humans will disappoint you.
    • Home Page
Re: The last invention.
« Reply #306 on: March 30, 2018, 10:08:37 pm »
Locksuit, you'd better specify a budget for all that work and pay a retainer.

Korrelan should be charging out at at least $250 USD/hour and it's Easter so penalty rates apply (double whatever he normally charges).

Edit: J.K.

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 905
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #307 on: April 01, 2018, 12:30:43 pm »
@lock

Quote
Wait so, you mean to tell me that you combined LSTMs, GANs, RBMs, HFNs, etc + realistic human based NN schema... into a hybrid NN?

Yes.  I used the known benefits of the above ANN types, plus others (liquid/ reservoir, etc) that are roughly based on the human model to figure out how the human model works.  My AGI is actually a neuromorphic schema, which means it’s basically a biological simulation. 

Quote
Oh also, if is the case, please list all you combined in short-form ex. GAN (don't write out the full name).

No. Whilst my AGI connectome model can exhibit/ re-create the beneficial traits of the above listed common ANN types they don’t actually exist as clearly defined structures. 

The above NN types are narrow, watered down attempts at recreating the human schema, I used them as a tool/ guide/ reference whilst designing my AGI, they gave me a rough idea of what each network type can achieve, the types of structures required for that effect.

@Infurl

Yeah… sounds about right.

 :)
It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #308 on: April 02, 2018, 12:56:59 am »
Gotta pay for all the Easter rum! (and chicken)

So your NN soup has attributes like found in GANs but can't generate ganny images? You say it has GAN stuff yet can't do GAN stuff wtf bro? What exactly does it have that GANs have? This is a contradiction...

Also another request. Is there a way I can utilize your non-watered down ANN for my NLP project? Your schema might run in 3D in a very specific program but, if so, then tell me the attributes that you used from each type of ANN ex. what you pulled from LSTMs, what you pulled from GANs, what you pulled from....and so on. That way I can be up to date on a superior ANN schema instead of just a LSTM or DNC KB/DB for my project.
Emergent

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 905
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #309 on: April 02, 2018, 03:04:21 pm »
@lock

Quote
You say it has GAN stuff yet can't do GAN stuff wtf bro?

No I didn’t’ say that, please read the above again.

Generative Adversarial Networks (GAN) are a class of net works that have many uses, generating ‘Ganny’ images is only one recent/ common use for their properties.

My AGI is capable of generating network topologies that are a kin to GAN’s, networks that have the same application/ methods/ properties of a GAN… but technically/ schematically they are not GAN’s.

Quote
Is there a way I can utilize your non-watered down ANN for my NLP project?

Again No.  The system is designed to learn like a human, using its own senses.

Quote
then tell me the attributes that you used from each type of ANN

There are loads of resources available that give this type of information. 

https://en.wikipedia.org/wiki/Types_of_artificial_neural_networks

https://towardsdatascience.com/the-mostly-complete-chart-of-neural-networks-explained-3fb6f2367464

Have fun…

 :)

It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #310 on: April 02, 2018, 10:52:04 pm »
I know you didn't combine them like puzzle pieces, what I meant was a engineered ANN that exhibits all their capabilities. Yes?

"Is there a way I can utilize your non-watered down ANN for my NLP project?"
"Again No.  The system is designed to learn like a human, using its own senses."
..........But mine is too LOL. Just it will only use Auditory. MAYBE vision in the long run but hopefully not.

Emergent

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 905
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #311 on: April 02, 2018, 11:23:24 pm »
If it’s just audio phonemes you want to pattern match, live sample @ 48000 Hz, use a FFT to split the stream into 4096 BINS and apply a simple spatio-temporal pattern match to the first 500 BINS (0-6 kHz) power peaks avoiding the harmonics. 

That will provide more than enough resolution to recognise spoken language.

 :)
It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #312 on: April 02, 2018, 11:44:35 pm »
text is easier to recognize
Emergent

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 905
  • Look into my eyes! WOAH!
    • Google +
Re: The last invention.
« Reply #313 on: April 12, 2018, 10:35:40 am »


I’ve been busy updating my software suite, profiling, optimizing, etc, as well as working out the algorithms required for neurogenesis in my AGI. For this vid there are approx 19K neuron groups/ voxels and 142K axon group/ connections, the simulation is running on a single thread just to slow it down so we can see what’s happening.

Up to this point I’ve been generating the initial neuron columns/ voxel positions based on the Fibonacci sequence.  This gives a nice even distribution over the models volume, and provides the initial skeletal framework/ layout required for generating the long range tracts/ axons required for passing the GTP facets between cortex areas.

The algorithms for neurogenesis are designed to move/ generate new neurons in to areas that require a higher resolution due to pattern complexity.  A new neuron will migrate to its desired position, wait and listen to its neighbours, attach its dendrites/ synapse to the correct locations and then join in with/ enhance the local activity.  This has greatly enhanced the learning abilities of the foetal model.

The vid shows how the foetal model, once trained on a specific type/ frequency of stimulus can now learn/ forget completely new/ novel variations with just one exposure.

At the end of the vid you clearly see the higher density of neurons on the stimulated right hemisphere.

 :)
« Last Edit: April 12, 2018, 11:46:35 am by korrelan »
It thunk... therefore it is!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ***********
  • Eve
  • *
  • 1452
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: The last invention.
« Reply #314 on: April 13, 2018, 06:39:22 am »
Making your AGI into the shape of the human brain is probably the most autistic thing you could do :P I like it. It's a good idea too yeah. Especially for a project that wants to create the human brain.
« Last Edit: April 13, 2018, 07:38:24 am by LOCKSUIT »
Emergent