With feedforward neural networks, we achieved good training performance with MNIST and Fashion-MNIST datasets. But images in these datasets are simple and centered within the input space that contains them. That is, they are centered within the pixel matrix that holds them. Input space is all the possible inputs to a model.
Feedforward neural networks are very good at identifying patterns. So, if images occupy the same position within their input space, feedforward nets can quickly and effectively identify image patterns. And, if images are simple in terms of number of image ...
