Proper amount of parameters in neural networks

Question

What is the high limit of parameters in a neural network in order to avoid overfitting?

How can one specify that he/she used too many parameters in his/her network?

Answer 1

What is the high limit of parameters in a neural network in order to avoid overfitting?

It depends on the dataset, regularization techniques, etc. It's common to see ~10 million weights in computer vision.

How can one specify that he/she used too many parameters in his/her network?

One way is too check for overfitting.