Page 2 of 2
Re: Alife and Machine Learning on the X16
Posted: Fri Jun 16, 2023 4:05 pm
by Ed Minchau
Well, the logistic function is another way of normalizing, but you then have to go full floating point. I've been trying to keep everything in the range [-1..1] to represent all numbers in eight bits.
Re: Alife and Machine Learning on the X16
Posted: Fri Jun 16, 2023 4:27 pm
by Ed Minchau
The big advantage of limiting neurons to only two inputs is that it turns neural computation from order of (N^2)logN operations into order of N operations for N neurons.
The second advantage is that it makes every neuron into something that can be displayed graphically as 3D: each input is one axis, the output is the third axis. So, the programmer has a much easier time visualizng what each neuron does.