ResNet readout_zero_init=True?

[D-X-Y]

Dear Greg,

Awesome project! May I ask why the linear output layer in ResNet is initialized as 0 instead of Gaussian(mean=0, var=1) as mentioned in the paper?

Thanks a lot for your time and help.

[thegregyang]

Thanks @D-X-Y!

We discuss this in the paper in Section D.2.

Let us know if you have any question after reading this.

[D-X-Y]

Thanks a lot for your explaination!