Parameter initialization
In order to assure a good fitting start, the model weights have to be initialized to the most effective values. Neural networks, which normally have a tanh activation function, are mainly sensitive to the range [-1,1], or [0,1]; for this reason, it's important to have the data normalized, and the parameters should also be within that range.
The model parameters should have useful initial values for the model to converge. One important decision at the start of training is the initialization values for the model parameters (commonly called weights). A canonical initial rule is not initializing variables at 0 because it prevents the models from optimizing, as they do not have a suitable function slope multiplier to adjust. ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access