Implement Papers

would be a good reference.

This looks very good. Should we take papers where code is written in Pytorch and meet one day in a week to discuss the paper?

Yes let us take already implemented papers in Pytorch and read and discuss the paper and write out understanding in the google docx. Can you please start a good doc link?

Do you guys have any paper in mind to start with? Which day of week and time is good to meet and discuss?

I want to start with the CycleGAN paper, what do you all think.

1 Like

Can we schedule a day in the week to discuss papers?

We have a weekly meetup, See details here.

This week I had presented Easy Data Augmentation Paper
We could discuss the papers during our next meetup. Would you like to lead the discussion for next week?

1 Like

Thanks I have nothing present next week. I just wanted to join a group of people who are interested in sharing ideas to implement papers.

No worries, Please feel free to jump on-board! :slight_smile:
Hopefully I’ll prepare for something, I’m still hunting for papers to present.

My goal is to join a group of people to discuss how maths equations are mapped to code in Pytorch in existing papers. And I want to do this enough time so that we can generalize this on a new paper.

1 Like

I would love to do the same!

My aim is the same as well. I think this will help me in learn code and math at the same pace. Nevertheless, I think we need a seperate forum just for math2code. The best example that I have come across is when jeremy explained WGAN in lesson 12 of part2 v2. Scroll to the wgan section of the notes. I find this very unique about fastai.


If a paper talks about a custom version of a generic architecture, usually they provide a diagram or a graph with details. I am working on my coding skills and trying to make it more readable. The following picture is taken form this tweet. The left depicts the xception architecture in a paper and the right one part depicts its tensorflow functional api code.

The picture below is taken from this tweet, where jeremy refactors the same thing in fastai.

3 Likes

Sounds great, I intend to do something on these lines. First, this makes the paper look less magical and more approachable. And often in the process of reverse engineering one comes up with many tweaks/variations. This is a better way to start, instead of waiting for that million dollar idea to descend into our consciousness!

2 Likes

I would love to join. I’ve been wanting to do the same as you.

I am also interested. Thanks for the suggestion.

I am also interested. I am also looking for same. Thanks

Just want to let you all know that we are meeting on Sunday morning 8 am PST for 30 min to 45 min. This meeting is not very formal in the beginning, so that everyone gets chance to participate and discuss ideas. As we learn more we will make this meeting more formal. You all are welcome to join…

Details? How to join meeting?

Am in.

Join the slack channel.

Hi I’m having difficulties understanding this section on the new paper by Ian Goodfellow entitled LAG(latent Adversarial Generator).

Link to the paper: https://arxiv.org/pdf/2003.02365.pdf