Personal understanding of convolutional neural network

Personal understanding of Convolutional Neural Networks

It's been almost a year since I entered school and now I have been exposed to deep learning , Today, I would like to summarize my understanding of convolution , There may be a lack of understanding , Welcome criticism ！
Convolutional neural networks are usually composed of many different middle layers , Including convolution layer 、 Pooling layer 、 Full connectivity layer, etc , Generally, convolutional neural network is used to train. When we give an input image, it passes through different network layers , The final output layer is obtained through forward propagation , Then calculate the difference between the predicted value and the real value as the error loss . Then the error is back propagated by deriving the loss , After each layer of neural network, the optimizer is used to update the model parameters, which completes an iteration , Usually, this process is repeated until the model converges .
What exactly is convolution ？ In fact, one sentence generalizes , Convolution layer is mainly used to extract features . The discovery of convolution is modeled on the human brain , When studying the human brain, biologists found that there are many small receptive domains on the neurons of the human visual center , Will respond to visual stimuli within the small receptive domain of vision . Inspired by this , Convolution also appears . There is also a receiving domain in convolution operation, that is, convolution kernel scans the picture back and forth , The acceptance domain of convolution kernel is overlapping , Together, they form a complete vision .