There is the "k means algorithm". And there is the kNN, or k nearest neighbor algorithm.
K-means is used on a image to find borders between patches of colors. That are used to create lines around them
as a edge detector does. So you can find the length of lines or contours and the area of blobs, and other metrics.
With the measurements that k-means generated, then kNN can find repeating features such as legs. And then kNN
can be used again to find grouping of features in a certain area, such as four, head two or one eye, body and a tail.
KNN is to find the dead center of a classification of the average thing that its looking for. like four legged animals
K means finds DATA border between horse and pony and cow. Using the features detected.
KNN can identify the cod, sardines, and blue water.
K means can find the VISUAL border between a school of cod, and the school of sardines, and the blue sea.
K means can operate as a unsupervised algorithm. But Knn is a supervised algorithm so a overseer must instruct
it what to detect.
Convolutional? Convolutional kernel as in a convolutional neural network?
Knn and k means are working at the same time automatically just by the give inherent design of neural networks.
But the final outputs is very much like KNN.
In the Brain. The brain is the overseer of Knn operations. It know kwap about the world in the beginning. its first
Knn move is just assume something exist. such as one color of a pixel and then goes out looks for it in
a image. The brain does this by making a scan kernel of on pixel that is 1 x 1 pixel matrix, made out of
nerves. the color selection is made by locking the randomly bouncing weight matrix into place. Nerves are
very noise when no used. if this scanning kernel detect something then it is keep. If not the weight are allowed to float
again.
With the successful detection, They are cloned and then mutated little and then checked to see if they exist.
When other colors have been detected then the brain assumes that to patches of color a next to each other or
spaced a certain distance between each other then goes out and look for it. if it finds it it is keep and then expanded on
by tame evolution mutation process. This is the k means process.
On the deeper layer of the brain combination of more complex arrangements of the simpler layer before it are made.
Very complicated detection can be made this way.
A detection device is not memory and only works at the moment it happens. So detection are stored into
a encoder neural network. The brain can give a hash code to detection and but they are give to
the most extreme of examples and other example that are between these
are give a distance value to the nearest extremes recorded hash value.
The brain can do massive parallel search searches and will have little lag in finding anything it is looking for.
