This question already has an answer here:
I have been trying to set up a ConvNet to classify some data. This data should be classified to either 1 (being what I need to get from the image) and 0 for everything that is irrelevant. I have successfully extracted 50k samples (positive) but I am having a hard time of getting negative samples. What would happen if I trained my net with 15k positive samples and lets say 5k negative ? I have read that this could be a problem for statistical algorithms ... is that also relevant to convolutional neural networks ?