A mechanical neural net

Hmm There's a lot of 'high falutin' words there, I'm not sure I understand. Is it something like an analog computer?

George H.

That's even goofier than an electrical neural net.

It all started with the mathematical model of the neuron, a very long time ago. I was most fortunate to read the right article and test the right software when was it? The nineties? the eighties? I have no reference at hand, but you could start reading this;

It is quite amazing what a simple artificial neural net can do, More and more these are used in every day life, from steering cars to medical diagnosis, image recognition, what not. I think it is as essential to know about this as it is to know about programming.

But I do think coming in from pure theory at this stage maybe be difficult. I started coding simple neural nets long ago... Have not done anything with it in years, but the principle is very clear to me. Bit like 'tronix. BTW there is a lot of software these day to play with that stuff. Been soldering all evening, part of the day, build some stuff, it actually works, invented a new circuit, more on that later. Lead...

That's too long for now... I got distracted by 'Turing's B-type machines'.

I know nothing of neural networks, is it like the program/computer that taught itself how to play go, and is now the world master?

That stuff is scary, someday a computer will teach itself to design circuits.

George H.

George Herold wrote

Computer were scary to some, cars, trains, were too. What I found interesting is that a configured (learned) neural net does not give a 'reason' for its decisions, like math has proof. It can be, just like us, 100% wrong. Some example was a car auto pilot wrongly interpreting some traffic sign,. it could be easily fooled.

These NN will become better, There was an article some time ago about a project by some prof who wanted to make the NN say _why_ it made a decision. Like you asking me 'why did you turn left there'. I could then answer 'it is a shortcut' or something. Very interesting, if it is part of the training (and becomes part of the weights) in the net, it could actually do that, he, that prof, had some success.

That would make it more easy for people to interact with it, different from 'this is the truth and live with it'

?? Next will be the religious neural net: 'Believe in my master or else'

Wow, you may well have a point.

On the bright side: lots of progress is being made with NNs, we do know very little about how the real neurons work, there are my types of those...

And there you are today: