# Neural networks backpropagation

I was reading chapter 2 of Michael Nielsen's book on deep learning:

Shouldn't the circle word read "forward" instead of "backward" because we are using the weighted inptut to layer l?

No, as the previous sentence, with "backward" italicized states, a measure of the error is moving from layer $$l + 1$$ back to layer $$l$$. This is reducing the layer index, i.e., going backwards among the numbered activation layers.