• vcmj@programming.dev
    link
    fedilink
    arrow-up
    1
    ·
    1 year 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.