[LOCKSUIT]

Why did Ranch say the only way to quickly search will be KNN, yet Convolutional is the way it actually works KNN is just a helper it doesn't find the features or nothing!

KNN is "smith" enters the brain, and he is sent to a room. The room is called memory. There's too many memory. He can't find himself/his clone copy. If KNN had all the memories stored so a mountain was near another poiny image and a wall was near another wall and a pin was near a pine needle, he could look over here, say nope, then be sent a mile away to a new type of images. Then, the brain says smith is my brother. The actions are spoken, senses are re-sensed/remembered.

Convolutional makes the input image pass all memory images but they are compressed small hashes, then the feature winner is chosen, KNN doesn't actually pick a winner all that good other than how much it "matches" by simply a match ratio system when it approaches a image to test. It takes longer. In fact the AIs can have trillions of years per second using the convolutional system - the input image divided into a few billion/enters a billion eyes, then a few atoms away each copy will pass the small hashes right there as a flat grid above the eye/eyes, then the winner is selected just after the flat array, and the linked actions to the winner are selected just a few atoms after from this point. KNN requires the signal to travel oooodles of atoms. Each atom wastes thousands of years - a photon makes millions of moves in a atom's space.

[keghn]

 ha look someone re invented Ranch's auto encoder, CAE:

https://pgaleone.eu/neural-networks/2016/11/24/convolutional-autoencoders/

[LOCKSUIT]

?...
Are you insulting Ranch?
Or mean that?

What I learned:
can learn specialized convo filters for certain input (links to filters) by filter #s being higher/lower (rid noise/blur & show white outlines on black contrast)

[keghn]

 Here is a demo of Ranches auto encoder, or CAE This way different then a CNN:

! No longer available

Auto encoder recreate the image or data that the are shown.


 KNN is a "bag of features"  it detects if all parts are in that area that make up a horse. Knn need other algorithms to define
the sub parts, like a edge detector algorithm. KNN does not say how they are connected to each other. Just that
they are there. A "bag of word"  is a grabbing of words. two bag can have the same words in them but when you look
at them they are completely different sentences.

When the editing distance and the physical distance to each sub part is add to KNN the it more like a
cnn. But these distance data is not connected to any of the sub features directly it just a bunch of data thrown into
a bag.

 KNN and CNN are very different but give the same answer. Butt CNN do it way better.

[LOCKSUIT]

Here is a demo of Ranches auto encoder, or CAE This way different then a CNN:

! No longer available

Auto encoder recreate the image or data that the are shown.


 KNN is a "bag of features"  it detects if all parts are in that area that make up a horse. Knn need other algorithms to define
the sub parts, like a edge detector algorithm. KNN does not say how they are connected to each other. Just that
they are there. A "bag of word"  is a grabbing of words. two bag can have the same words in them but when you look
at them they are completely different sentences.

When the editing distance and the physical distance to each sub part is add to KNN the it more like a
cnn. But these distance data is not connected to any of the sub features directly it just a bunch of data thrown into
a bag.

 KNN and CNN are very different but give the same answer. Butt CNN do it way better.

All I got out of that is:

KNN the it more like a
cnn.

it just a bunch of data thrown into

a bag.

KNN and CNN are very different but give the same answer.

Butt CNN do it way better.

..............................................................................Slowly, slipping, away. . .

I'm out.






I might figure it out this weekend.