[Study Notes] Week1_Convolutional Neural Networks_Why Convolutions?

1. The advantages of using convolutional layers:

    1) Parameter sharing

        Feature detectors used in one part of the image (such as vertical edge detection) tend to apply to another part of the image, greatly reducing the amount of parameters

        

    2) Sparse connection

        Each output value is only related to a small part of the input value, preventing overfitting

2. CNN still has Cost Function (cost function) and reduces Cost through optimization algorithms

    

Guess you like

Origin http://43.154.161.224:23101/article/api/json?id=325523979&siteId=291194637