enas-pytorch
enas-pytorch copied to clipboard
Any trained result that match the paper?
Hi,
Is there any trained result that match the paper?
Best, zmonoid
+1
I trained 150 epoch using default config, and got 58.8% must worser than 68.0% from enas excepted performance
In this implementation, aux_head is not used. This could be a potential reason
I trained 150 epoch using default config, and got 58.8% must worser than 68.0% from enas excepted performance
In this implementation, aux_head is not used. This could be a potential reason
I read some paper published recently , it seems that ENAS is a little unstable, the revolution of AutoMl is far more than accomplished .
Currently I don't have the computation resource to reproduce the result. But in my previous experiment, the final result (training the searched architecture using the official TensorFlow implementation) was on par with the paper.
I not really understand the implementation of weight sharing.Could you please give me some tips?