/sci/ - Science & Math

[Anonymous No.7067996]

In the Wikipedia article on deep learning, the subsection discussing deep neural networks mentions that a DNN can be trained using backpropagation with stochastic gradient descent, updating weights using the equation

<span class="math">\Delta w_{ij}(t+1) = \Delta w_{ij}(t) + \eta \frac{\partial C}{\partial w_{ij}}[/spoiler]

Why does this equation use addition rather than subtraction?

[Anonymous No.7067997]

http://en.wikipedia.org/wiki/Deep_learning#Deep_neural_networks

is the section in question.

[Anonymous No.7067999]

>>7067996
holy shit man sauce on that gif

[Anonymous No.7068030]

because they felt like having a negative learning rate. you can do it either way.