About normalising data

So Jeremy said In Introduction to NN that normalized data works better. Well, I understand the formula but I’m still when you normalize & when not?

Also any practical example of what can happen we if normalize vs if we don’t

1 Like

I think you should normalize your data in all cases. I am not quite sure what the theory normalizing your data is, but I guess it is following the approach to make it as easy as possible for your NN to learn, so it might be harder to train (take longer) your NN if you would omitted the normalization step.

Normalization is used to make your data easier comparable, that means your data is converted so that it lives on the same scale.

Take e.g. images with values 0-100 and others with values 110-210, e.g. caused by different lighting conditions. These are not comparable. After normalization they are.

1 Like

what do mean by different lighting condition & how maths x-m/s fix lighting conditions?

Can you explain the explain more? Clear it?