I have recently enjoyed watching Jeremy Howard’s interview in Lex Fridman Artificial Intelligence (AI) Podcast. Totally worth it if you haven’t seen it.
At time 55:15 (youtube link below) in the interview, Jeremy says: “We’ve actually recently shown you don’t need GANs. We have developed GAN level outcomes without needing GANs”
He also mentions these techniques being related to transfer learning and new loss functions.
I’m very interested about this topic and I would appreciate if anyone in the community could point me to papers, code or discussions around this topic.
Thank you for your help,