hiromis / notes Goto Github PK

Hi Hiromi
First of all big thank you for the work you did and effort you put into making the notes. They are so detailed and helped so many students to learn Deep Learning. I've seen them used everywhere!

Just a suggestion. What if the notes where in jupyter notebook format so that the incorporated code can be run for example at Google colab, kaggle kernels or locally? Jupyter files can be viewed at github anyway if someone don't want to run the code, just read.

just an idea...

Keep up the good work!

Hi! There's a little error with the formula presented in RMSProp. The formula has x^2 but it should be \sqrt{x}.

Thanks for making this! :)

I'm a student, and I am learning these lessions. I have finished L1~L5. So would you QQ(470137285) me the URL? Thanks!

thanks for your notes which help me a lot.
This is the most detailed notes of the course I've ever seen.
i want to know whether you will finish notes in part 2.

I'm looking forward to your response.
@hiromis

Hello.

Thank you very much for the notes.

Im not sure why we are cloning in the featureloss function :

out_feat = self.make_features(target, clone=True)
as the vgg used for the loss is frozen anyway ?
Thank you in advance.

There can be up to a thousand layers of neural network. ResNet34 has 34 layers, and ResNet50 has 59 layers, but let's look at layer one.

IMHO it's just a typo. ResNet50 has 50 layers?

Thank you for the great notes.

The Putting your model into production section of Lesson 2 notes describes Simon Willison's [cougar-or-not] (https://github.com/simonw/cougar-or-not) web app deployment using Zeit. It looks like Zeit no longer supports this type of deployment. The article has been removed from the fast.ai course, so the https://course-v3.fast.ai/deployment_zeit.html link is broken and should be removed, and perhaps the description of Zeit for hosting should be updated as well.

@hiromis Many thanks for sharing your excellent notes!