Amazing lecture @jeremy. Looking forward to learn about how to create our own dataset from google images. Thanks!
gotta have label, this is Supervised training. Unsupervised training is possible, but that’s in Part 2.
if we do nn.parallel what different would it do
Hi Jeremy, can I rewatch the todays lecture? Its been during the night here in EU, and I was not as awake as I wanted to be …
runs the model on multiple-gpus
Learning rate is too high. So parameter updates are inadequate. Neatly explained here https://towardsdatascience.com/estimating-optimal-learning-rate-for-a-deep-neural-network-ce32f2556ce0
No problem So is it something like what @KevinB said, that this number is actually the prediction percentage for the incorrect class (and not the correct one like Jeremy said) ? But if it’s not I really don’t understand this point…
Great session, also a huge thanks to the people answering questions live on here!
is this resolved anytime soon or can i do some custom changes so it work for me on my colab fastai functions
If I’m not mistaken, the video will be available right away after the lecture
No, idea, I was just linking to where it’s discussed, please continue over there
Thanks Jeremy and Rachel, excellent lecture.
the learn.lr_find() finds the optimal rate for which layers?
Pay attention to the fact that ResNet-50 takes less around half of the operations to achieve equally competitive results. So though I agree that right now Inception-v4 is the best if you want to be very accurate, but ResNet wins in terms of efficiency.
Any threads that have instructions for using multiple GPU with fastai library?
I searched, but not the one that with instructions.
Thank you in advance!
Thanks a lot!
search for nn.DataParallel