Dear @a-non-e-moose

I just "spoke" to @neavvy (as we already got to know each other in real life and I really adore this boy) and he pointed out that your knowledge is mindblowing. And he surely had a point :)

Amazing comment!

ps.
I noticed you don't have much SP. Would you like me to support you with small 40sp delegation? It would allow you to stay a bit more active without reaching resource credit limits.

Cheers
Piotr

Thank you for your reply @a-non-e-moose. Appreciate that.

> what type of function is f(x,y)? I presuppose it's a multiplication, but I see no bias in your perceptron.

I decided not to include bias on my simple scheme, as it would most probably confuse many readers of this article. In fact you guessed, the third input of my function is bias.

> Is it a sigmoid function, a heaviside step function or something else entirely? Is the output a boolean? I don't imagine you use back-propagation to train this simple of a network, so how was the activation function modified?

Well, using all of those sophisticated methods would be indeed clueless for such non complicated model. So I decided to use simple multiplication algorithm. Output is boolean (true makes the bird jump, false makes the bird continue falling) and the activation function is pretty simple: if the sum of inputs multiplied by their weights is positive output is true, otherwise it is false.

> Perhaps it was reinforcement training, but the concurrent multiple trials suggest to me that it may be an evolutionary algorithm.

The training was reinforcement. I decided to use a bigger number of birds in order to increase the probability of training a perfect player faster. But all of those birds were completely independent.

> No need to respond here, especially if your upcoming article will address all these questions. I can wait for that. I just thought a more detailed set of questions may lend itself more easily to being concisely answered.

I am always happy to respond :) Especially as you seem to share very similar passions with me.