Scoring function:
The picture is stretched into a one-dimensional matrix x, that is, 3072x1, the final 10 classification of the score value is 10x1, that W will have to be 10x3072 matrix, that is, 10 sets of 3,072 characteristics of the weight value, multiplied by x, plus B, to get a 10x1 matrix, This matrix is the final scoring value for each category.
Assuming that the image is divided into 2x2 pixels, and finally 3 type, then the image can be stretched to 4x1 matrix, the final result is the 3x1 matrix, then the weight w can only be 4x3 matrix.
That is, there are 3 sets of weight parameters, each set of parameters have 4 characteristics of the weight, where the 3 weights of the parameters corresponding to the last category of 3 categories, 4 characteristics of the weight of each feature is the importance of the category (where negative value indicates a reaction).
In this example, the last to get the dog scored the highest value. This is, of course, an example and does not mean universality.
--------------------------October 14, 2018 02:52:49--sleep---------------------------------------
?
002-Neural Network Foundation