• @vcmj
    link
    18 months ago

    Or I should explain better: most training samples will be cut off at the top, so the network sort of learns to ignore it a bit.