Normalizing (Normalization) in order to eliminate the dimension between the different data to facilitate data comparison and co-processing, such as neural networks, normalization can speed up the convergence of network training;
1. decimal between data becomes (0,1), primarily for the convenience of data processing, because the data is mapped to the range of 0 to 1, the process can be made more convenient and fast.
2. dimensionless expression transformed into dimensionless expression become a scalar. After normalization of data, in the same order of magnitude, you can eliminate the effects of dimension and dimensionless unit between indicators, improve data comparability between different indicators.
3. Main Algorithm: linear conversion, i.e. normalization min-max (conventional method)
y = (x-min) / (max-min)
Standardization (Standarlization) is to facilitate the next step in data processing, and data conversion scaling, etc., not to facilitate the processing or compare with other data together, such as the mean standardized data after a 0-1, more conducive to the use of the standard normal distribution too properties, for processing;
Regularization (Regularization) using prior knowledge, thus introducing regularization (Regulator), increasing role in guiding constraint processing, such as used in the regularized logistic regression, can effectively reduce the over-fitting phenomenon.