bug in 2nd order?

[Interesting6]

I see in your code, just using self.net(x_spt[i], fast_weights, bn_training=True)

however the torch.autograd.grad() method contain the following parameter:

create_graph (bool, optional) – If True, graph of the derivative will be constructed, allowing to compute higher order derivative products. Default: False.

Is that means your code just calculates the 1st order derivative?

Thank you!

[jayzhan211]

self.net is not using torch.autograd.grad()

[jingjingjing-666]

same confusion

[Vampire-Vx]

I think it is a 1st order approximation actually

[iamxiaoyubei]

+1