Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

Yes, and now I extend it to audio (beat detection) on something like this:
https://botcompany.de/images/1102653

On Thu, Sep 5, 2019, 12:10  wrote:

> I think your segmentation is different to what im thinking.   ur doing
> vision recognition with it.
> *Artificial General Intelligence List *
> / AGI / see discussions  +
> participants  + delivery
> options  Permalink
> 
>

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M8d197f5dd3be0b1c3e0c02db
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[rouncer81]

I think your segmentation is different to what im thinking.   ur doing vision 
recognition with it.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-Ma62ee3bd4885a44422fb1971
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

Yeah but how should a machine know which filters to apply? Therein lies the
art

On Thu, 29 Aug 2019 at 14:39,  wrote:

> if you have a 3d camera, segmentation is even easier. its not even
> really machine learning,  its just filters.
> *Artificial General Intelligence List *
> / AGI / see discussions  +
> participants  + delivery
> options  Permalink
> 
>


-- 
Stefan Reich
BotCompany.de // Java-based operating systems

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M85b5c5866d0bfdbfffdb5088
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[rouncer81]

Yep.  labels first,  actual understanding later on.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-Mcb4fa9e98e050133e8e97495
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[immortal . discoveries]

I agree rouncer81, the visual cortex classifies objects like words first, then 
it will see them next to each other ex. car>road. This is the higher "sentence" 
temporal network. You don't recognize them as a joined object, but rather parts 
next to parts that make up a part next to another part repeat.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M62200c06aaf75ff2058e92dd
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[rouncer81]

Matt Mahoney I have an argument against that,  computer vision ends before 
symbolic relations start.

Your saying that your eye invents jokes with what it sees,  I say no,  the 
vision just classifies the visible aspect alone.
The rest of the derivation of the eye is the rest of the brain.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M6d5c0fd241ff61dd9717585a
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Matt Mahoney]

I doubt segmentation will help with image recognition. You lose context.
You recognize people not just by their faces but by when and where you see
them, who they are with, and what they say. It is easier to recognize a car
on a road than a car or a road on a white background.

We tried word segmentation in speech recognition and parsing in sentence
recognition. It doesn't work very well.

On Thu, Aug 29, 2019, 8:39 AM  wrote:

> if you have a 3d camera, segmentation is even easier. its not even
> really machine learning,  its just filters.
> *Artificial General Intelligence List *
> / AGI / see discussions  +
> participants  + delivery
> options  Permalink
> 
>

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M56375d2d7d10ab11a9a4b53e
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[rouncer81]

Good to here u still sound like your on top of things.

I think theres no need for bad confidence,  theres going to be a simple 
solution for this singularity business.
--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M4820fde7dd51af5fadb5e5d5
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

And you KNOW how much I love neural networks. (Seems you haven't heard of
my stance yet.)

On Fri, 23 Aug 2019 at 04:37, Secretary of Trades 
wrote:

> Intros to "mixed" from MiniRem's department
>
> https://arxiv.org/pdf/1804.01452.pdf
>
> https://arxiv.org/pdf/1904.09013.pdf
>
>
> On 20.08.2019 23:06, Stefan Reich via AGI wrote:
> > Random forests are interesting... but I usually just don't have number
> > inputs. All I have is text
> >
> > On Tue, 20 Aug 2019 at 22:05, Secretary of Trades
> > mailto:costi.dumitre...@gmx.com>> wrote:
> >
> > When the turk squares the apple and inference does no better
> > there's no
> > crop, only crap...
> >
> >
> > With that... a mixed data db, even when the associations are more
> than
> > binary, might require a certain kind of ANN, modeled so that it
> > resembles a 'random forest' (growing the tree at discrete times -
> with
> > or without the Sleep States) and also a reactor when propagating the
> > inference particles. It takes a large correspondence table before
> > building or packing the model.
> >
> >
> > https://en.wikipedia.org/wiki/Random_forest
> >
> >
> > On 20.08.2019 21:17, Stefan Reich via AGI wrote:
> > > lol... "crap ANNs". You convinced me there...
> > >
> > > On Tue, 20 Aug 2019 at 00:04, Secretary of Trades
> > > mailto:costi.dumitre...@gmx.com>
> >  > >> wrote:
> > > :)
> > >
> > > It's annotation (a text entry). From Amazon's "mechanical
> turk".
> > >
> > > A hilarious way of referring the current norm of primitive and
> > > supervised learning molded on crap ANNs.
> > >
> > >
> > > On 20.08.2019 00:11, Stefan Reich via AGI wrote:
> > > > I find your explanation hard to follow. What is "turking"?
> > Is this
> > > > another one of those "nothing must be manual because we are
> > > making AI"
> > > > argument? Those are really useless.
> > > >
> > > > On Mon, 19 Aug 2019 at 22:36, Secretary of Trades
> > > >  >   > >
> > >  > 
> > >  >  > > >
> > > > Hearing, but not for control. For segment cuts to be
> > > associated with
> > > > image cuts (e.g. 'red numbers').
> > > >
> > > > Also, the red numbers should not be sourced by any kind
> of
> > > turking
> > > > (such
> > > > as the color channel threshold kind of is). Instead a
> hard
> > > or soft
> > > > focusing should source the image portion of interest.
> > > >
> > > > As one would bring an apple in front of a child's eyes
> and
> > > say "Apple"
> > > > or "An apple" ...
> > > >
> > > > Such a set-up would be unsupervised enough (no text in
> the
> > > > database and
> > > > no other kind of turking/ annotation), but instead an
> > > association
> > > > between an image db (where non-OCRed text can be
> included)
> > > and a sound
> > > > cuts db (where spoken text could also be included). The
> > > learning model
> > > > can be complex enough so that a spoken phrase that
> > describes
> > > the apple
> > > > fruit to be linked (or associated) not to an image but to
> > > only a tiny
> > > > segment of the same spoken phrase (the 'apple' noun, when
> > > isolated).
> > > >
> > > > So, instead of sourcing for control, the hearing
> > module should
> > > > source to
> > > > a smart association algorithm that runs continuously
> > (so that
> > > > associations can be made at discrete times).
> > > >
> > > >
> > > > On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> > > > > You mean voice control? Yeah I have code for that.
> > > > >
> > > > > On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> > > > >  > 
> > >  > >
> > 
> > >  > >>
> > > >  > 
> > >  > >
> > > >  >   

Re: [agi] You can help train desktop image segmentation

[Secretary of Trades]

Intros to "mixed" from MiniRem's department

https://arxiv.org/pdf/1804.01452.pdf

https://arxiv.org/pdf/1904.09013.pdf


On 20.08.2019 23:06, Stefan Reich via AGI wrote:

Random forests are interesting... but I usually just don't have number
inputs. All I have is text

On Tue, 20 Aug 2019 at 22:05, Secretary of Trades
mailto:costi.dumitre...@gmx.com>> wrote:

When the turk squares the apple and inference does no better
there's no
crop, only crap...


With that... a mixed data db, even when the associations are more than
binary, might require a certain kind of ANN, modeled so that it
resembles a 'random forest' (growing the tree at discrete times - with
or without the Sleep States) and also a reactor when propagating the
inference particles. It takes a large correspondence table before
building or packing the model.


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


On 20.08.2019 21:17, Stefan Reich via AGI wrote:
> lol... "crap ANNs". You convinced me there...
>
> On Tue, 20 Aug 2019 at 00:04, Secretary of Trades
> mailto:costi.dumitre...@gmx.com>
>> wrote:
> :)
>
> It's annotation (a text entry). From Amazon's "mechanical turk".
>
> A hilarious way of referring the current norm of primitive and
> supervised learning molded on crap ANNs.
>
>
> On 20.08.2019 00:11, Stefan Reich via AGI wrote:
> > I find your explanation hard to follow. What is "turking"?
Is this
> > another one of those "nothing must be manual because we are
> making AI"
> > argument? Those are really useless.
> >
> > On Mon, 19 Aug 2019 at 22:36, Secretary of Trades
> > mailto:costi.dumitre...@gmx.com> >
> 
>  >
> > Hearing, but not for control. For segment cuts to be
> associated with
> > image cuts (e.g. 'red numbers').
> >
> > Also, the red numbers should not be sourced by any kind of
> turking
> > (such
> > as the color channel threshold kind of is). Instead a hard
> or soft
> > focusing should source the image portion of interest.
> >
> > As one would bring an apple in front of a child's eyes and
> say "Apple"
> > or "An apple" ...
> >
> > Such a set-up would be unsupervised enough (no text in the
> > database and
> > no other kind of turking/ annotation), but instead an
> association
> > between an image db (where non-OCRed text can be included)
> and a sound
> > cuts db (where spoken text could also be included). The
> learning model
> > can be complex enough so that a spoken phrase that
describes
> the apple
> > fruit to be linked (or associated) not to an image but to
> only a tiny
> > segment of the same spoken phrase (the 'apple' noun, when
> isolated).
> >
> > So, instead of sourcing for control, the hearing
module should
> > source to
> > a smart association algorithm that runs continuously
(so that
> > associations can be made at discrete times).
> >
> >
> > On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> > > You mean voice control? Yeah I have code for that.
> > >
> > > On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> > > mailto:costi.dumitre...@gmx.com>
> >

> >>
> > 
> >
> > 
>  wrote:
> > >
> > > It sees. It must also hear.
> > >
> > >
> > > On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > > > https://www.youtube.com/watch?v=W6eJSCGEkSE
> > > >
> > > > --
> > > > Stefan Reich
> > > > BotCompany.de // Java-based operating systems
> > > > *Artificial General Intelligence List
> > >

Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

Random forests are interesting... but I usually just don't have number
inputs. All I have is text

On Tue, 20 Aug 2019 at 22:05, Secretary of Trades 
wrote:

> When the turk squares the apple and inference does no better there's no
> crop, only crap...
>
>
> With that... a mixed data db, even when the associations are more than
> binary, might require a certain kind of ANN, modeled so that it
> resembles a 'random forest' (growing the tree at discrete times - with
> or without the Sleep States) and also a reactor when propagating the
> inference particles. It takes a large correspondence table before
> building or packing the model.
>
>
> https://en.wikipedia.org/wiki/Random_forest
>
>
> On 20.08.2019 21:17, Stefan Reich via AGI wrote:
> > lol... "crap ANNs". You convinced me there...
> >
> > On Tue, 20 Aug 2019 at 00:04, Secretary of Trades
> > mailto:costi.dumitre...@gmx.com>> wrote:
> >
> > :)
> >
> > It's annotation (a text entry). From Amazon's "mechanical turk".
> >
> > A hilarious way of referring the current norm of primitive and
> > supervised learning molded on crap ANNs.
> >
> >
> > On 20.08.2019 00:11, Stefan Reich via AGI wrote:
> > > I find your explanation hard to follow. What is "turking"? Is this
> > > another one of those "nothing must be manual because we are
> > making AI"
> > > argument? Those are really useless.
> > >
> > > On Mon, 19 Aug 2019 at 22:36, Secretary of Trades
> > > mailto:costi.dumitre...@gmx.com>
> >  > >> wrote:
> > >
> > > Hearing, but not for control. For segment cuts to be
> > associated with
> > > image cuts (e.g. 'red numbers').
> > >
> > > Also, the red numbers should not be sourced by any kind of
> > turking
> > > (such
> > > as the color channel threshold kind of is). Instead a hard
> > or soft
> > > focusing should source the image portion of interest.
> > >
> > > As one would bring an apple in front of a child's eyes and
> > say "Apple"
> > > or "An apple" ...
> > >
> > > Such a set-up would be unsupervised enough (no text in the
> > > database and
> > > no other kind of turking/ annotation), but instead an
> > association
> > > between an image db (where non-OCRed text can be included)
> > and a sound
> > > cuts db (where spoken text could also be included). The
> > learning model
> > > can be complex enough so that a spoken phrase that describes
> > the apple
> > > fruit to be linked (or associated) not to an image but to
> > only a tiny
> > > segment of the same spoken phrase (the 'apple' noun, when
> > isolated).
> > >
> > > So, instead of sourcing for control, the hearing module should
> > > source to
> > > a smart association algorithm that runs continuously (so that
> > > associations can be made at discrete times).
> > >
> > >
> > > On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> > > > You mean voice control? Yeah I have code for that.
> > > >
> > > > On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> > > >  >   > >
> > >  > 
> > >  >  > > >
> > > > It sees. It must also hear.
> > > >
> > > >
> > > > On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > > > > https://www.youtube.com/watch?v=W6eJSCGEkSE
> > > > >
> > > > > --
> > > > > Stefan Reich
> > > > > BotCompany.de // Java-based operating systems
> > > > > *Artificial General Intelligence List
> > > > > * / AGI / see
> > discussions
> > > > >  + participants
> > > > >  +
> delivery
> > > options
> > > > > 
> > Permalink
> > <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M1e8cdc19650b4a48d80aaa6c
> > > >
> > > >
> > > > --
> > > > Stefan Reich
> > > > BotCompany.de // Java-based operating systems
> > > > *Artificial General Intelligence List
> > > > * / AGI / see discussions
> > > >  + participants
> > > >  + delivery
> > options
> > > > 

Re: [agi] You can help train desktop image segmentation

[Secretary of Trades]

When the turk squares the apple and inference does no better there's no
crop, only crap...


With that... a mixed data db, even when the associations are more than
binary, might require a certain kind of ANN, modeled so that it
resembles a 'random forest' (growing the tree at discrete times - with
or without the Sleep States) and also a reactor when propagating the
inference particles. It takes a large correspondence table before
building or packing the model.


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


On 20.08.2019 21:17, Stefan Reich via AGI wrote:

lol... "crap ANNs". You convinced me there...

On Tue, 20 Aug 2019 at 00:04, Secretary of Trades
mailto:costi.dumitre...@gmx.com>> wrote:

:)

It's annotation (a text entry). From Amazon's "mechanical turk".

A hilarious way of referring the current norm of primitive and
supervised learning molded on crap ANNs.


On 20.08.2019 00:11, Stefan Reich via AGI wrote:
> I find your explanation hard to follow. What is "turking"? Is this
> another one of those "nothing must be manual because we are
making AI"
> argument? Those are really useless.
>
> On Mon, 19 Aug 2019 at 22:36, Secretary of Trades
> mailto:costi.dumitre...@gmx.com>
>> wrote:
>
> Hearing, but not for control. For segment cuts to be
associated with
> image cuts (e.g. 'red numbers').
>
> Also, the red numbers should not be sourced by any kind of
turking
> (such
> as the color channel threshold kind of is). Instead a hard
or soft
> focusing should source the image portion of interest.
>
> As one would bring an apple in front of a child's eyes and
say "Apple"
> or "An apple" ...
>
> Such a set-up would be unsupervised enough (no text in the
> database and
> no other kind of turking/ annotation), but instead an
association
> between an image db (where non-OCRed text can be included)
and a sound
> cuts db (where spoken text could also be included). The
learning model
> can be complex enough so that a spoken phrase that describes
the apple
> fruit to be linked (or associated) not to an image but to
only a tiny
> segment of the same spoken phrase (the 'apple' noun, when
isolated).
>
> So, instead of sourcing for control, the hearing module should
> source to
> a smart association algorithm that runs continuously (so that
> associations can be made at discrete times).
>
>
> On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> > You mean voice control? Yeah I have code for that.
> >
> > On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> > mailto:costi.dumitre...@gmx.com> >
> 
>  >
> > It sees. It must also hear.
> >
> >
> > On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > > https://www.youtube.com/watch?v=W6eJSCGEkSE
> > >
> > > --
> > > Stefan Reich
> > > BotCompany.de // Java-based operating systems
> > > *Artificial General Intelligence List
> > > * / AGI / see
discussions
> > >  + participants
> > >  + delivery
> options
> > > 
Permalink


> >
> >
> > --
> > Stefan Reich
> > BotCompany.de // Java-based operating systems
> > *Artificial General Intelligence List
> > * / AGI / see discussions
> >  + participants
> >  + delivery
options
> >  Permalink
> >


>
> --
> Stefan Reich
> BotCompany.de // Java-based operating systems
> *Artificial General Intelligence List
> * / AGI / see discussions
>  + participants
>  + delivery options
>  Permalink
>

Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

lol... "crap ANNs". You convinced me there...

On Tue, 20 Aug 2019 at 00:04, Secretary of Trades 
wrote:

> :)
>
> It's annotation (a text entry). From Amazon's "mechanical turk".
>
> A hilarious way of referring the current norm of primitive and
> supervised learning molded on crap ANNs.
>
>
> On 20.08.2019 00:11, Stefan Reich via AGI wrote:
> > I find your explanation hard to follow. What is "turking"? Is this
> > another one of those "nothing must be manual because we are making AI"
> > argument? Those are really useless.
> >
> > On Mon, 19 Aug 2019 at 22:36, Secretary of Trades
> > mailto:costi.dumitre...@gmx.com>> wrote:
> >
> > Hearing, but not for control. For segment cuts to be associated with
> > image cuts (e.g. 'red numbers').
> >
> > Also, the red numbers should not be sourced by any kind of turking
> > (such
> > as the color channel threshold kind of is). Instead a hard or soft
> > focusing should source the image portion of interest.
> >
> > As one would bring an apple in front of a child's eyes and say
> "Apple"
> > or "An apple" ...
> >
> > Such a set-up would be unsupervised enough (no text in the
> > database and
> > no other kind of turking/ annotation), but instead an association
> > between an image db (where non-OCRed text can be included) and a
> sound
> > cuts db (where spoken text could also be included). The learning
> model
> > can be complex enough so that a spoken phrase that describes the
> apple
> > fruit to be linked (or associated) not to an image but to only a tiny
> > segment of the same spoken phrase (the 'apple' noun, when isolated).
> >
> > So, instead of sourcing for control, the hearing module should
> > source to
> > a smart association algorithm that runs continuously (so that
> > associations can be made at discrete times).
> >
> >
> > On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> > > You mean voice control? Yeah I have code for that.
> > >
> > > On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> > > mailto:costi.dumitre...@gmx.com>
> >  > >> wrote:
> > >
> > > It sees. It must also hear.
> > >
> > >
> > > On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > > > https://www.youtube.com/watch?v=W6eJSCGEkSE
> > > >
> > > > --
> > > > Stefan Reich
> > > > BotCompany.de // Java-based operating systems
> > > > *Artificial General Intelligence List
> > > > * / AGI / see discussions
> > > >  + participants
> > > >  + delivery
> > options
> > > >  Permalink
> > > >
> >  <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M1e8cdc19650b4a48d80aaa6c
> > >
> > >
> > > --
> > > Stefan Reich
> > > BotCompany.de // Java-based operating systems
> > > *Artificial General Intelligence List
> > > * / AGI / see discussions
> > >  + participants
> > >  + delivery options
> > >  Permalink
> > >
> > <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M5cdb99ac21573fb433315836
> >
> >
> > --
> > Stefan Reich
> > BotCompany.de // Java-based operating systems
> > *Artificial General Intelligence List
> > * / AGI / see discussions
> >  + participants
> >  + delivery options
> >  Permalink
> > <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-Ma3202fbd050a786e8256b1c6


-- 
Stefan Reich
BotCompany.de // Java-based operating systems

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M81891c7bd381ee1a2effcac9
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Secretary of Trades]

:)

It's annotation (a text entry). From Amazon's "mechanical turk".

A hilarious way of referring the current norm of primitive and
supervised learning molded on crap ANNs.


On 20.08.2019 00:11, Stefan Reich via AGI wrote:

I find your explanation hard to follow. What is "turking"? Is this
another one of those "nothing must be manual because we are making AI"
argument? Those are really useless.

On Mon, 19 Aug 2019 at 22:36, Secretary of Trades
mailto:costi.dumitre...@gmx.com>> wrote:

Hearing, but not for control. For segment cuts to be associated with
image cuts (e.g. 'red numbers').

Also, the red numbers should not be sourced by any kind of turking
(such
as the color channel threshold kind of is). Instead a hard or soft
focusing should source the image portion of interest.

As one would bring an apple in front of a child's eyes and say "Apple"
or "An apple" ...

Such a set-up would be unsupervised enough (no text in the
database and
no other kind of turking/ annotation), but instead an association
between an image db (where non-OCRed text can be included) and a sound
cuts db (where spoken text could also be included). The learning model
can be complex enough so that a spoken phrase that describes the apple
fruit to be linked (or associated) not to an image but to only a tiny
segment of the same spoken phrase (the 'apple' noun, when isolated).

So, instead of sourcing for control, the hearing module should
source to
a smart association algorithm that runs continuously (so that
associations can be made at discrete times).


On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> You mean voice control? Yeah I have code for that.
>
> On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> mailto:costi.dumitre...@gmx.com>
>> wrote:
>
> It sees. It must also hear.
>
>
> On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > https://www.youtube.com/watch?v=W6eJSCGEkSE
> >
> > --
> > Stefan Reich
> > BotCompany.de // Java-based operating systems
> > *Artificial General Intelligence List
> > * / AGI / see discussions
> >  + participants
> >  + delivery
options
> >  Permalink
> >
 

>
>
> --
> Stefan Reich
> BotCompany.de // Java-based operating systems
> *Artificial General Intelligence List
> * / AGI / see discussions
>  + participants
>  + delivery options
>  Permalink
>


--
Stefan Reich
BotCompany.de // Java-based operating systems
*Artificial General Intelligence List
* / AGI / see discussions
 + participants
 + delivery options
 Permalink



--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M3f29ca3069184417d3da6522
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

I find your explanation hard to follow. What is "turking"? Is this another
one of those "nothing must be manual because we are making AI" argument?
Those are really useless.

On Mon, 19 Aug 2019 at 22:36, Secretary of Trades 
wrote:

> Hearing, but not for control. For segment cuts to be associated with
> image cuts (e.g. 'red numbers').
>
> Also, the red numbers should not be sourced by any kind of turking (such
> as the color channel threshold kind of is). Instead a hard or soft
> focusing should source the image portion of interest.
>
> As one would bring an apple in front of a child's eyes and say "Apple"
> or "An apple" ...
>
> Such a set-up would be unsupervised enough (no text in the database and
> no other kind of turking/ annotation), but instead an association
> between an image db (where non-OCRed text can be included) and a sound
> cuts db (where spoken text could also be included). The learning model
> can be complex enough so that a spoken phrase that describes the apple
> fruit to be linked (or associated) not to an image but to only a tiny
> segment of the same spoken phrase (the 'apple' noun, when isolated).
>
> So, instead of sourcing for control, the hearing module should source to
> a smart association algorithm that runs continuously (so that
> associations can be made at discrete times).
>
>
> On 19.08.2019 15:37, Stefan Reich via AGI wrote:
> > You mean voice control? Yeah I have code for that.
> >
> > On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
> > mailto:costi.dumitre...@gmx.com>> wrote:
> >
> > It sees. It must also hear.
> >
> >
> > On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > > https://www.youtube.com/watch?v=W6eJSCGEkSE
> > >
> > > --
> > > Stefan Reich
> > > BotCompany.de // Java-based operating systems
> > > *Artificial General Intelligence List
> > > * / AGI / see discussions
> > >  + participants
> > >  + delivery options
> > >  Permalink
> > >
> > <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M1e8cdc19650b4a48d80aaa6c
> >
> >
> >
> > --
> > Stefan Reich
> > BotCompany.de // Java-based operating systems
> > *Artificial General Intelligence List
> > * / AGI / see discussions
> >  + participants
> >  + delivery options
> >  Permalink
> > <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M5cdb99ac21573fb433315836


-- 
Stefan Reich
BotCompany.de // Java-based operating systems

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-Ma3202fbd050a786e8256b1c6
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Secretary of Trades]

Hearing, but not for control. For segment cuts to be associated with
image cuts (e.g. 'red numbers').

Also, the red numbers should not be sourced by any kind of turking (such
as the color channel threshold kind of is). Instead a hard or soft
focusing should source the image portion of interest.

As one would bring an apple in front of a child's eyes and say "Apple"
or "An apple" ...

Such a set-up would be unsupervised enough (no text in the database and
no other kind of turking/ annotation), but instead an association
between an image db (where non-OCRed text can be included) and a sound
cuts db (where spoken text could also be included). The learning model
can be complex enough so that a spoken phrase that describes the apple
fruit to be linked (or associated) not to an image but to only a tiny
segment of the same spoken phrase (the 'apple' noun, when isolated).

So, instead of sourcing for control, the hearing module should source to
a smart association algorithm that runs continuously (so that
associations can be made at discrete times).


On 19.08.2019 15:37, Stefan Reich via AGI wrote:

You mean voice control? Yeah I have code for that.

On Mon, 19 Aug 2019 at 05:54, Secretary of Trades
mailto:costi.dumitre...@gmx.com>> wrote:

It sees. It must also hear.


On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> https://www.youtube.com/watch?v=W6eJSCGEkSE
>
> --
> Stefan Reich
> BotCompany.de // Java-based operating systems
> *Artificial General Intelligence List
> * / AGI / see discussions
>  + participants
>  + delivery options
>  Permalink
>




--
Stefan Reich
BotCompany.de // Java-based operating systems
*Artificial General Intelligence List
* / AGI / see discussions
 + participants
 + delivery options
 Permalink



--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M228cdb08a1febd9ebe7307b5
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

You mean voice control? Yeah I have code for that.

On Mon, 19 Aug 2019 at 05:54, Secretary of Trades 
wrote:

> It sees. It must also hear.
>
>
> On 19.08.2019 01:31, Stefan Reich via AGI wrote:
> > https://www.youtube.com/watch?v=W6eJSCGEkSE
> >
> > --
> > Stefan Reich
> > BotCompany.de // Java-based operating systems
> > *Artificial General Intelligence List
> > * / AGI / see discussions
> >  + participants
> >  + delivery options
> >  Permalink
> > <
> https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M1e8cdc19650b4a48d80aaa6c
> >
> >


-- 
Stefan Reich
BotCompany.de // Java-based operating systems

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M5cdb99ac21573fb433315836
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Secretary of Trades]

It sees. It must also hear.


On 19.08.2019 01:31, Stefan Reich via AGI wrote:

https://www.youtube.com/watch?v=W6eJSCGEkSE

--
Stefan Reich
BotCompany.de // Java-based operating systems
*Artificial General Intelligence List
* / AGI / see discussions
 + participants
 + delivery options
 Permalink




--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-Me73a7d903d7096e96b7a2e31
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Re: [agi] You can help train desktop image segmentation

[Secretary of Trades]

Huawei is looking for an OS


On 19.08.2019 01:31, Stefan Reich via AGI wrote:

https://www.youtube.com/watch?v=W6eJSCGEkSE

--
Stefan Reich
BotCompany.de // Java-based operating systems
*Artificial General Intelligence List
* / AGI / see discussions
 + participants
 + delivery options
 Permalink




--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-Mf2f18b396fb188918211f7a2
Delivery options: https://agi.topicbox.com/groups/agi/subscription

[agi] You can help train desktop image segmentation

[Stefan Reich via AGI]

https://www.youtube.com/watch?v=W6eJSCGEkSE

-- 
Stefan Reich
BotCompany.de // Java-based operating systems

--
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8f7f05f86e62415a-M1e8cdc19650b4a48d80aaa6c
Delivery options: https://agi.topicbox.com/groups/agi/subscription