For example if you have a segmentation or regression problem, but the features that you are interested in are always in the same or similar place in the input image or time series (across different training/test samples), then does it some how make more sense to use a locally connected input layer rather than a convolutional layer because the features that you want to learn will be specific to that location, enabling a shallower network with fewer parameters ?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.