2 points | by impossiblefork 8 hours ago ago
1 comments
I saw this neat paper, which proposes a more practical way to train models whose output are the fixed point of a learned function and thought it would be reasonable to talk about.
They use the trick from Feistel ciphers for reversibility.
I saw this neat paper, which proposes a more practical way to train models whose output are the fixed point of a learned function and thought it would be reasonable to talk about.
They use the trick from Feistel ciphers for reversibility.