Hi,
I have been attempting to run the code for the CTR method.
The Convolution layers are randomly causing NaNs sometimes during the forward pass for valid inputs. Have you observed this and any suggestions on how to overcome? SHould the Nans be reset to some specific informed values, for example?
Hi,
I have been attempting to run the code for the CTR method.
The Convolution layers are randomly causing NaNs sometimes during the forward pass for valid inputs. Have you observed this and any suggestions on how to overcome? SHould the Nans be reset to some specific informed values, for example?