Regularization vs Decreasing Units Manually

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

If regularization makes our network act more like a linear function, can we achieve a similar effect by decreasing the number of units in our hidden layers manually? When would I choose one over the other? And why isn’t “smaller network” in the flow chart options for solutions to high variance?

2

It is an interesting question! Here’s another recent thread on the same topic.

1 Like