Three lines two neurons and M×M receptors.

[MacIntoShiba]

Dear all, I have lately been interested in AI.

One of my thoughts has been as follows:

What if we have two 'neurons' and the world consists of vertical, horizontal and diagonal lines. How would we go about making sure only one neuron is assigned to either vertical or horizontal lines and that diagonal lines are neglected. Another related question would be the minimum amount of visual receptors which provide inputs to these two neurons. Let's say we need M×M recptors to spot a line. With M needs at least be 3×3, but maybe 4×4 would give better results. So there is a tradeoff between cost (number of receptors) and reward, certainty with which lines can be detected.Also I would like to start reading up on statistics/probability/neural networks. I have not studied these topics before so it needs to be an introduction. I am also a bit allergic to books with 1000+ pages so shorter is better. I have done a master in mechanical engineering,
so I do know about lin algebra, calculus etc.

 

[chiro]

Hey MacIntoShiba.

There are a number of books on the areas you are seeking and for probability and statistics you will first have to learn probability and after that do statistical inference, regression and applied probability.

First books on probability often look at the same thing and something like below is a good introduction:

https://www.amazon.com/dp/0495110817/?tag=pfamazon01-20

After that you look at statistical inference, regression analysis and applied probability.

Here is a good applied probability model book:

https://www.amazon.com/dp/0124079482/?tag=pfamazon01-20

Graduate stuff goes into more abstraction and into different data types along with a more rigorous treatment of probability.

If you have specific questions on probability and statistics let us know in the thread so we can help you.