Forward propagation of deep learning neural networks (1)

Method

Calculate the score first , Then enlarge the difference , Finally, normalization is carried out

• Calculation of score , Calculate by multiplying the weight matrix with each pixel .
• Expand differences , adopt ex function
  
• The last part softmax Normalize the function of
• 

Calculation of loss function

• Through the difference between the correct classification and each wrong classification + 1 The lion Finally, find the average , You can get the wrong classification score of each picture
• The formula , The score calculated for the weight parameter
• 
• softmax function
  

Regular process sorting

Get the task , Divide a picture into three categories , Cat, dog and pig

Initialize the weight parameter randomly

• For example, the image size is 32*32
  

After calculating the score , Expand differences

• utilize e^x function
  
  For example, the score is ,12, 32, 2
  After expansion
  28, 322, 4

normalization

Calculate the loss function

• loss = -logx
  
• Add - Because all the numbers are negative , stay y The bottom half of the shaft

Update weight parameters in reverse