Filters used in conv2D


From Week 1, programming assessment I understand that we are not designing the filter parameters manually and will be updated during training.

Then is it possible to see, which features were learned by a filter after training the network, as we are not designing the filter parameters manually? is it possible to access the filter parameter tensor data?


Hey @siddhesh_mane,
In the Week 4 of this course, there is a video entitled “What are deep ConvNets learning?” In that, Prof Andrew highlights and describes a research work which does this exact thing. You may take a look into it first if you want to, otherwise, you may reach it in a sequential order as well. I leave it up to you.