Weights and activation function

Question

How are the weights decided?

How is it decided what kind of activation function I use?

Is loss function the same as activation function? If not , could there be different loss functions?

Answer 1

Weights are the parameters which are optimized during the training process. So we do not decide them, we compute them.

The activation function is something to try. Now we have converged to a set of reliable activations functions but technically it's trial and error.

The loss function is a different concept so no. There are multiple loss functions as well depending on your objectif (MSE, or MAE for regression for instance, Cross-entropy for classification)

Comments

Thank you for the explanation.
What are the abbreviations in
MSE, or MAE for regression for instance, Cross-entropy for classification)
please?

---

Mean Square Error, Mean Absolute Error.