Open caixincx opened 2 years ago
Hello! In paper, the loss formulated as -log(\tau(p^t)), but I found that in the code is -log(p^t). Is it the reason why it cannot reach the acc?
I tried the $-\log(\tau(p^t))$ too, but the performance is still poor. I think there may be other tricks that should be considered.
Hello! In paper, the loss formulated as -log(\tau(p^t)), but I found that in the code is -log(p^t). Is it the reason why it cannot reach the acc?