[Kaeldric]

After a month of searching and studying articles and books chapters I finally understand the inner meaning of a Boltzmann machine ... and it is amazing!
It's incredible how a physic formula bounding the state of the probability of a set of  particles and their energy can be fitted into abstract idea of a neural network leading to a so important mathematical structure!
And, above all, I very love the sensation you feel when, at last, you understand how something like that works. There is still some formulas I must still "digest" ... but I will do soon.
Sorry for the useless post ... but I had to tell someone

[ivan.moony]

The harder the goal is, more thrilled you feel when you reach it

[keghn]

do you mean restricted Boltzmann machine?

[Kaeldric]

No. I wanted to understand the basic concepts behind the full version. I will study the restricted one later.

[keghn]

 Any of this going on within sound round Boltzmann?

New Theory Cracks Open the Black Box of Deep Learning:

https://www.quantamagazine.org/new-theory-cracks-open-the-black-box-of-deep-learning-20170921/ 

[ranch vermin]

I bet they have a beautiful output,   because they are so god darn computational to run!

It depends on the symantics and rules you put into them how good they go as a robot brain tho.

[keghn]

 I would love to see a tutorial for Boltzmann machine. Hinton and Larosh give the worst tutorial ever. A tutorial for
for some who very new to it is very much needed. Simple tutorial for the beginner with out the math would be great. I have been looking
for good Boltzmann and a restricted Boltzmann for years and they are all crwap!

[Kaeldric]

Here is try to explain Boltzmann machines in "english".
It helped me a little when I started ... I hope it can help you too.

http://rocknrollnerd.github.io/ml/2015/07/18/general-boltzmann-machines.html

[keghn]

 Thanks @Kaeldric. 
 Thank for the great info but i am forever cursed in not understanding BM and RBM's
 All other network i get. I got my highly hacked version of RBM that is nothing like real RBM, in my attempt to understand it. But i
i think i can live with out them. I noticed on all other forum that when people asked about them a sudden dead silence comes
from a noise audience.  I even heard Andrew Ng say something about he rather use something else then a RBM.
 I specialize in unsupervised learning. I rather use some other than a RBM. 

[keghn]

 My  Hacked RBM has two layers. The front is the input and back is the output. The font layer
would be 1000 x 1000 and so would be the other layer.
 The way it work is a 24 x 24 image of a letter or a number would be inputted on the lower left
bottom corner. All other first layers do not take inputs. 
  Ready steady go!
 The image bounce between the two layers. As it does it shifts the image toward the upper right corner. And erosion  of the image occurs.
 When it become a single dot it stops, then that is it's clustering position on the out put side. Curved edges of the symbol
 and  the corners dissolve very fastest and push it left or right. Straighter edges dissolve very slowly.


 My Notes: 
Learning a Manifold of Fonts:
http://vecg.cs.ucl.ac.uk/Projects/projects_fonts/projects_fonts.html 


In Raw Numpy: t-SNE: 
https://nlml.github.io/in-raw-numpy/in-raw-numpy-t-sne/



Ricci Flow - Numberphile: 
 

[keghn]

Deep Learning Library 1.0 - Fast Neural Network Library: 

https://baptiste-wicht.com/posts/2017/10/deep-learning-library-10-fast-neural-network-library.html