I am trying to optimize the learning rate of a neural net and found that the precision of the optimize operator is much less than the precision of the parameter itself. The parameter description allows values in the range from 4.9E-324 to 1. If I put that into the optimize parameters it will only generate parameters starting from 0.000 to 0.001 which might not be sufficient for this parameter. I think the precision is limited which is quite unfortunate. The question is, Is that a bug or am I doing something wrong? Please find the demo process attached.