The number of parameters in aneural net as used today, specifically in computer vision, is basically never learned from training data. I actually cannot recall

The number of parameters in aneural net as used today, specifically in computer vision, is basically never learned from training data. I actually cannot recall practically used methods that would do that.