machine learning - Convolutional neural networks vs downsampling? -

after reading on subject don't understand: is 'convolution' in neural networks comparable simple downsampling or "sharpening" function?

can break term down simple, understandable image/analogy?

edit: rephrase after 1st answer: can pooling understood downsampling of weight matrices?

convolutional neural network family of models proved empirically work great when comes image recognition. point of view - cnn different downsampling.

but in framework used in cnn design there comparable downsampling technique. understand - have understand how cnn works. build hierarchical number of layers , @ every layer have set of trainable kernels output has dimension similiar spatial size of input images.

this might serious problem - output such layer might extremely huge (~ nr_of_kernels * size_of_kernel_output) make computations intractable. reason why techniques used in order decrease size of output:

1. stride, pad , kernel size manipulation: setting these values value decrese size of output (on other hand - may lose of important information).
2. pooling operation: pooling operation in instead of passing output layer outputs kernels - might pass specific aggregated statistics it. considered extremely useful , used in cnn design.

for detailed description might visit tutorial.

edit: yes, pooling kind of downsampling 😊