CODEWITHSUNDEEP
neural-network
user2170547
user2170547

Reputation: 361

Zero input in neural network

The fellowing image is a neural network version of xor I found at wikipedia: enter image description here

If I try to imagine how it might work in code it's clear with the inputs 0/1, 1/0 because the last purple neuron is firing a 1 in both cases. But what happened for 0/0 and 1/1?

For 0/0 the first neurons aren't firing at all and for 1/1 it's the purple neuron that doesn't fires. So the network gots "stuck" there.

But how does one know if the network gots "stuck" or if it's just not ready calculating? Maybe I imagine the thing to asynchronous? Thanks!

Upvotes: 0

Views: 600

Answers (1)

el.pescado - нет войне
el.pescado - нет войне

Reputation: 19236

I think you are confused by "no firing" concept. When neuron inputs are below threshold, it doesn't "wait" for that threshold, it propagates 0. So, it's not that neuron "doesn't fire", stopping processing, it just "zero signal" to further neurons.

For 0/0 case, red neurons don't fire, so green neurons get 0 on input, so they don't fire either, so purple one gets 1*0 + -2*0 + 1*0 = 0, so it doesn't fire, so you get 0 on output. So, as you can see, even if some neuron doesn't fire, you process signals further.

For 1/1 case, red neurons fore, so do green ones, so purple one gets 1*1 + -2*1 + 1*1 = 0 which doesn't fire, so 0 is returned.

Upvotes: 1

Related Questions