Why does not Generative Adversarial Networks use bias in convolutional layers?

[alexbellgrande (Alex Bratt)]

I wondered this initially too. I think it’s because the beta term in batch norm effectively adds a bias to each channel.