Deep Internal Learning: Deep Learning from a Single Input

arXiv:2312.07425v2 Announce Type: replace
Abstract: Deep learning, in general, focuses on training a neural network from large labeled datasets. Yet, in many cases there is value in training a network just from the input at hand. This is particularly relevant in many signal and image processing problems where training data is scarce and diversity is large on the one hand, and on the other, there is a lot of structure in the data that can be exploited. Using this information …

abstract arxiv cases cs.cv cs.lg data datasets deep learning eess.iv eess.sp general image image processing network neural network processing signal training training data type value