This slide demonstrates what was said in the beginning about the filters - they look like the pattern they are trying to predict.
hdd
After iterating over a number of layers, due to the learning of weights and tuning by all the neurons, each neuron would respond to one feature ( as in it has learnt that particular feature). So when you do this in a sequence of a number of layers, we can see the end result of learning from all the weights aggregated in a fully connected layer. This layer has learnt to recognize the features that make a particular image and assigns labels based on its training for the subsequent test dataset
o_o
As you keep adding more and more layers, the filters keep pooling at one feature from the image. Thus, we have have neural networks that start learning about what an image is by continuously layering filters to find a feature.
This slide demonstrates what was said in the beginning about the filters - they look like the pattern they are trying to predict.
After iterating over a number of layers, due to the learning of weights and tuning by all the neurons, each neuron would respond to one feature ( as in it has learnt that particular feature). So when you do this in a sequence of a number of layers, we can see the end result of learning from all the weights aggregated in a fully connected layer. This layer has learnt to recognize the features that make a particular image and assigns labels based on its training for the subsequent test dataset
As you keep adding more and more layers, the filters keep pooling at one feature from the image. Thus, we have have neural networks that start learning about what an image is by continuously layering filters to find a feature.