Lecture 5 - Data Augmentation, Regularization and ResNets

Livestream Link: https://www.youtube.com/watch?v=sJF6PiAjE1M

Lecture Timing: June 20 (Sat), 8.30 AM PST/9:00 PM IST

Notebooks & References:

What to do after the lecture?

Asking/Answering Questions :
Reply on this thread to ask questions during and after the lecture. Before asking, scroll through the thread and check if your question (or a similar one) is already present. If yes, just like it. During the lecture, we’ll answer 8-10 questions with the most likes. The rest will be answered on the forum. If you see a question you know the answer to, please post your answer as a reply to that question. Let’s help each other learn!


@aakashns There seems to be a conflict in some of the emails you sent out for assignment deadlines. In the email you sent 20 minutes ago, you said that the deadlines for assignment 1 and 2 are now closed. However, in an email you sent on Saturday June 13th, you specified that the deadline for both assignment 2 and 3 was changed to “June 20th 9:00 PM IST/3:30 PM GMT”. It still hasn’t become 3:30 PM GMT, so I’m not sure why you’ve said that assignment 2 is now closed. Could you please explain if this is a mistake?

You have time, the link for assignment 2 is still open for 16 hours or so.


What’s the best approach to apply different learning rates to early layers using PyTorch? How to unfreeze the last layers when using transfer learning?

How to add extra features to input, for eg adding age and gender along with images in skin cancer detection?

@aakashns @PrajwalPrashanth can the submission deadline for the course project and kaggle competition be extended? The course was about learning GANs and some people ( including me) would like to make a project on GANs. and since GANs will be taught next week. So it will be great if we get 1 week after the lecture to work on GANs


Hey I have submitted all 3 assignments but have not received any PASS/FAIL messages only confirmations of assignments being submitted. Thanks in advance.

Will you guys acknowledge our kaggle submission?

1 Like

@PrajwalPrashanth I’m currently doing my project by implementing DeViSe on Pytorch, is that acceptable for the course project or do I need to do only classification/regression problem?

Assignment 5 - Course Project This is request of some other people mentioned in Assignment 5 thread link.

@aakashns @PrajwalPrashanth

Hi dear Jovian Team,

Do we have to do 1 submission per day for the competition?

Thanks in advance,



@aakashns, is there something you would like to tell us? xD

1 Like

When I uploaded my model to the data competition, the score was 0.71. When I checked the score on the site, it changed to 0.17. Why that happened?

How to choose the best pre-trained model for our problem?
How do we know which loss function and optimizers are better for our problem?

1 Like

Can You give an intuition how Augmentation actually works ? If training set has X number of images, will 2X / 3X different images generated when Aug/Transforms are applied?

@PrajwalPrashanth @aakashns The ‘Run on Kaggle’ button seems to run the notebook in the existing tab as well as on a new tab, Seems like a probable bug with the recent upgrades on the platform.

@aakashns @PrajwalPrashanth Hii guys. I have submitted all the previous assignments but haven’t got any pass or fail grade mail. I hope you look forward to it. So even I get a fail grade, I would have enough time to make a submission once again.

@aakashns Could you tell how to save best model/ model checkpoints during training in this leacture?

You can check out for differential learning implementations. Also fastai has a implementation too for reference.

You will have to customize layers to add new features. Things can get complex, Since CNNs is for images.

You can start with something like encoder-decoder CNN, where you can get a tensor representation of the image. Then you can add features to that. Please review about all of them, to proceed further.