Part 2 2018 videos?


I was wondering whether the new Part 2 videos were already available somewhere as a “pre-release”. Really looking forward to it!


Here too, looking forward for them!

(Gokul) #3

They are already available under each In-class thread.
Look for Part 2 In class thread
For Class 10 :


Hi @EpiphanyMania, thanks for the link. Unfortunately I don’t seem to have access to that topic. Is there a link to a youtube video?

(Gokul) #5

Unfortunately, if you don’t have access, you have to wait for the release of part 2 on Jeremy’s youtube channel.


@jeremy @rachel

Would you consider sharing pre-release videos here on the forum?

:fast_forward: :robot: <-- logo?

(Cedric Chee) #7

The in-person course has started on Mar 19 and will be available online a few months after completion. Use the 2017 version for now:

Edit: The 2018 version will be available in early July.


Hi @cedric, thanks for your reply. Yes, I saw that this was the plan.

However, it seems like a shame to wait a few months, because the half-life of this information is so short! A few months is an eternity in deep learning. :timer_clock::timer_clock::timer_clock:

(Cedric Chee) #9

Not really. If you understand’s teaching philosophy and approach:

It’s better to learn time-tested concepts and knowledge instead of chasing the non-time tested cutting-edge research like Reinforcement Learning. So, I guess what we study in Part 2 v1 (2017) won’t be so much different with what they will cover in the Part 2 v2 (2018) course. The most, the content for each key concepts will be refreshed, for e.g., instead of WGAN we will learn CycleGAN. A brief summary of what we’ll learn in each lesson (I extracted this from the 2017 version ppt):

  • Generative models
    • CNNs beyond classification:
      • Improve images
        • artistic style transfer
        • colorization
        • super-resolution
        • captioning
      • GANs
      • Localization
        • Segmentation
        • Detection
    • NLP beyond classification:
      • translation
      • Seq2Seq
      • Attention
      • Memory networks
  • Large data sets
    • Large images
    • Lots of data points
    • Large outputs


Hi @cedric. I agree somewhat on the concepts side, but for example GANs have moved on tremendously in one year. The bigger difference is on the practical side, everything changes, even the basic software framework was changed from Keras to PyTorch.

Anyway, I was wondering about the reason to delay publication until several months later. Will there be significant editing of the material? Or is it a legal requirement of USF?

(Cedric Chee) #11

I agree on the practical side :grin:

If my maths is correct, the in-person course will end on Apr 30. The 2018 version will be available in early Jul. So, 2 months later :thinking: I hope the 2 months timeline is for the official launch and they “pre-release” it sooner. But, in the past, this is the usual pattern. The reason why I have to think about this is because our study group for AI Saturdays have to decide which version to use.

This goes beyond me.

(Jason McGhee) #12

From what I understand, it is a combination of preparing the fast ai library to fully support the course material (clean up, organization, bug fixes), editing videos, cleaning up notebooks, etc.

There could be something with University policy, but that I do not know.