3 ms·
Nice demonstration, but misses to explain the bias value in the forward propagation step. While quite important, this value is often left out when demonstrating
by Chilinot 8y ago
Nice demonstration, but misses to explain the bias value in the forward propagation step. While quite important, this value is often left out when demonstrating the propagation function. So having it in warrants a short description in my opinion.
It also skips over the bias value in the back propagation step.
- ironSkillet 8y agoThe bias is just another input node whose value happens to be constant (granted with full connectivity), right? So the motivating idea/derivation of backpropogation doesn't change.
- Chilinot 8y agoThe bias is often updated/corrected in the back propagation step as well. The purpose of the value is to shift the activation function along the x-axis basically. While the weights define the slope of the activation function.