Neuron Update Rule "Binary" and Color

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view

Neuron Update Rule "Binary" and Color

I have decided to use Simbrain as my own Teach Yourself ANNs in 5 Minutes... .  Please excuse if my question is dumb!

I am setting up to create some AND/OR etc. gates. I (think I) get what is going on with colors in the default case with Linear updates rules.

So now I want to play with logic gate rules.  I have no inhibitory behaviour, I expect all neurons to have either 0 or 1 as value.

I set all my synapses to:

Strength: 0.5
Update Rule: Static

I set all my neurons to:

Increment: 1
Input Type: Weighted
Update Rule: Binary
On Value: 1
Off value: 0
Bias: 0

So my neurons have a value of 0 or 1, and each has 2 incoming synapses with value 0 or 0.5 to contribute. I can set a computed neuron's Threshold to 0.4 or 0.9 for OR/AND (I read somewhere your Threshold test is > rather than >=, shame 'coz it makes them less readable than 0.5 or 1.0 would be :)  Oh, I guess I could add a bias of 0.1 if I wanted that?)

I type most of the above in case anyone wants to say i am going about it the wrong way.  But my question is: when I change a neuron's Update Rule from Linear to Binary, when its value is 1 it now shows in "pale" red rather than "strong" red.  I have to manually set it to 2 if I want to see the "strong" red it used to have.  (On the other hand, if I set it manually to -1 it shows in "strong" blue, there is no "pale" blue value now.)

Why is this?  What is it telling me --- I'm confused?  I expected value 1 to be "full" red, as 1 is the maximum value.  This seems to be the case on a Binary neuron even if no synapses are defined, so I think it's neuron stand-alone behaviour.

Thanks :)

Reply | Threaded
Open this post in threaded view

Re: Neuron Update Rule "Binary" and Color

Cool!  You might consider an increment of .1 on the input modes just for playing with your logic gate networks.

For the behavior of binary neurons

You are right that it's > rather than >=.   Maybe we should add a check box allowing the >= behavior, if people want it.     I'd be open to that.

For the color of the "on" / "off" states in a binary neuron, I did it that way since I just the colors were a bit more appealing with a bit of saturation, but you're right that it does not quite fit expectations.  Here again I'm open to changing that, or adding an option    

In both cases, if you are others feel it's worth changing, submit a feature request (or pull request) here

- Jeff