How do I join?| Club Policy | Resources | Important Links | Surveys and Quizzes | Implementation Guidelines
From now on the updates will be posted on our website only! Please keep that in mind! LINK
Generative Adversarial Networks
Soft Deadline (paper): 23:59, July, 1st, 2019 (all timezones)
Hard Deadline (paper): 23:59, July, 4th, 2019 (all timezones)
Soft Deadline (Quiz): 23:59, July, 4th, 2019 (all timezones)
Hard Deadline (Quiz): 23:59, July, 8th, 2019 (all timezones)
Paper: Differentially Private Federated Learning: A Client Level Perspective
From now on the updates will be posted on our website only! Please keep that in mind! LINK
This is maybe the last post in this repository, because today I created a simple static website for the club (Github Pages + Jekyll), which simply looks nicer. It's not finished yet and there are a lot of small things to do, but you can take a look at it already. :) And yes, today, I don't share any updates about the papers. I was a bit (a lot, tbh) demotivated by what's going on with the challenge course and in the end stopped caring at all. I didn't post 60 days challenge update and not going to do it tomorrow or the day after tomorrow. Sticking to the things I really enjoy. And taking part in a race of who can be online every single day and do whatever weird thing is required is not one of them. Soooo today I simply decided to relax and made a nicer page for us. I will still continue with the club, but for the sake of the club, not the challenge course. :)
-- Cheers, Ana
There was a lot of confusion today on Slack regarding current paper, discussions, etc. So I updated our calendar to reflect all the current deadlines. I will be updating it as we add new deadlines. If I don't do so, please ping me. :) Also, note that each event has a link (e.g. to a quiz or to a paper), so you don't have to look for it. Hope you like it. The link - CALENDAR
-- Cheers, Ana
The next paper has been chosen and I'm pretty excitied about it. 35.14% of our respondents selected GANs. The second best choice was the paper on GNNs that gathered 16.22% of voices. There were total 37 votes, so I consider the selection phase complete and close the voting. The updates will follow soon!
Live discussion for the current paper (on differential privacy) will take place on Sunday on Slack (SPAIC Slack) at 5PM UTC. Don't forget to take the quiz if you haven't done so already.
-- Cheers, Ana
Aaaand, here goes THE QUIZ. It should be much easier than the previous one. Do enjoy! And again - we don't track any personal scores, only aggregated rating, so use it to enhance your reading experience!
-- Cheers, Ana
It's time to select the next paper! Looking at the list, I literally want to read that all, but the majority decides, so please VOTE! :)
Our second quiz will be published tomorrow - we're working on finalising it with Marce and Dharmendra. Don't worry - it will be an easy one this time. It will test only general understanding and will not go into too much detail. Specifically, don't worry about the calculation of epsilon and delta - there was discussion on Slack regarding this issue and maybe we'll post a summary later. But this requires studying another paper which in turn required studying another papers, so we don't include the details of calculation of epsilon/delta or details of moments accountant into the quiz. Good luck!
-- Cheers, Ana
There are just a few days remaining until the soft deadline for the current paper and I think most of you have already completed it. It's not that long after all. So for something completely different I wanted to share with you one of my favourite mock papers - I hope you can appreciate the humor. :) It's from SIGBOVIK (if you don't know what it is, do google). The paper is [Meta-meta-learning for Neural Architecture Search through arXiv Descent](https://www.bayeswatch.com/unofficial/meta-meta-learning.pdf). Do enjoy and let us proceed with arXiv descent... :)
-- Cheers, Ana
Quick update for today. You all (scholars!) know about about the challenge "60 days of Udacity". The activities you perform in our club also count, so if you worked on reading paper, completing the quiz, writing implementaiton or summary, DO list it as a activity in your daily report for the challenge. [IMPORTANT] We are now trying to arrange it with Akshit to make it clear what counts and what doesn't count for that challenge. We think that reading papers the way we did requires a lot of time and effort and should be counted. We will keep you posted.
-- Cheers, Ana
Hello everyone! There were no news for the past few days, because I have been at the event. By the way, you should definitely check it out - it's called Talent Meets Bertelsmann. Bertelsmann now is really into data and is looking for talent in that field. They will also sponsor about 50,000 scholarships with Udacity in the next 3 years, so look out for them! And I personally would like to stress that this is really a great company - there are so many divisions, so many interesting projects going on and so many opportunities. So if you have a chance and you think you have the required skills - apply next year!
On that note - I would like to announce the new deadlines. There is no quiz YET, so the deadlines are only for reading paper so far. Our current paper is quite short, so I expect that we complete it by the end of the week. Good luck!
-- From Berlin with Love, Ana
The survey is over and we have a winner - Differentially Private Federated Learning: A Client Level Perspective. In fact, it was a tie: Neural Turing Machine, word2vec and the winning paper all got almost equal number of votes. But since the challenge course is on privacy and federated learning is one of its topics, we decided that we should read this paper anyway if it gets enough voices. The new deadlines will be added tomorrow, but expect to have 2 weeks to complete the full cycle.
Update for Slack users of Secure and Private AI scholarship challenge: me, Marce and Dharmendra are now the moderators of #reading_paper_discuss channel! There are still more than two months ahead and I hope it will be a great journey!
-- Cheers, Ana & Marce
Hey everyone! After running a survey we discovered that majority of people WOULD LOVE to have live discussion. And there are great news for them! This Sunday (as most of you selected Saturday and Sunday as preferred day) there will be live session to discuss the paper we’ve just completed (A Tutorial on PCA). The session will be hosted by @shivu who completed all the stages (read -> quiz -> implement). This is a great time to discuss the toughest parts as well as practical applications. This is NOT ama, it’s discussion. @shivu and @Marce will be moderating the discussion, but they are not teachers, so prepare for the discussion, not recitation! Also, if there are other people who completed the full cycle (or at least reading and quiz), they can co-host the session if they wish - please message one of us (me, @Marce, @shivu or @Dharmendra Choudhary). Happy learning!
-- Cheers, Ana
Marce is running a survey to select our next paper. Please take the SURVEY. We're also planning live discussion on Sunday that will be lead by one of the members.
If you're still reading the paper and planning to take the quiz - don't panic! I will not close the quiz. The deadline is for self-control to make sure we're pretty much on the same page. But I will like the quiz open, so that newcomers can take it with their own pace.
-- Cheers, Ana
As promised, the quiz solutions are out! Please do not read before taking the quiz. It's important to think about the question before knowing the answer. ;)
The hard deadline for the paper has passed and the hard deadline for the quiz is tomorrow. What it means is that we will be moving to the next stage. It's time to select another paper and time to discuss what we've learnt. Also, it would be great if some of you will ry implementaton (see implementation guidelines).
Don't forget to vote, when the survey for the next paper is out, so that you don't ask me "Why are we reading this?" You know what I want to read, but we have democracy here, so voting decided it all. :)
-- Cheers, Ana
We have a lot of updates news today!
First, there is channel now on Challenge Course Slack. (those who are not a part of the challenge course, ignore the channel part).
Second, detailed answers to the quiz will be published tomorrow!
Third, there is a repo now for our first paper - feel free to add your implementations (or links to your implementations) there. Please read the implementation guidelines first. And there is already one implementation by @shivu!
And finally Tomorrow we will start the new round of voting for the next paper! Time to move on from PCA! Add your proposals to Paper Proposals sheet here
-- Cheers, Ana
There are 42 (forty two) members now (as per Google Doc) and it's getting hard to manage it all. Hence, I asked Dharmendra (@Dharmendra Choudhary) and Marce (@Marce) for help. They will help with posting the updates and managing other important issues. They are already doing that on Slack! But there are more great things to come! ;)
Also, I was told that a lot of members were struggling with the math part. Initially, I planned the club for those with maths and CS background, but eventually more and more people got interested, so this requirement was kind of removed. What's done is done and everyone is welcome! However, there are some ideas on how to adress this issue. Look for updates!
The first deadline has passed! Only a few people were able to complete the paper and the quiz by the deadline. Also, based on data, I suspect that some of them tried the quiz before completed the paper - otherwise, I can't explain the answers selected for some of the questions. :) I have initiated discussion in the #general channel, so you can join or start your own - just make sure to prepend it with the keyword papers_dclub!
-- Cheers, Ana
And our first quiz is up and running! The link: QUIZ
It has been tested for sanity by two of our dear members: Dharmendra and Marce. Thank you, guys!
Important things about this quiz:
- I will not see your individual answers, only the overall stats, so feel free to make mistakes. :)
- The quiz covers general concepts, not any specific parts of derivations. It should be accessible for all members!
I will publish the overall stats after all deadlines and maybe some intermediate results if I have enough data. Again - these are aggregate stats, no individual results are seen to anyone. But please avoid taking the quiz without reading the paper (or at least skimming through it).
Also, after you complete it, please take a survey so that we know how to proceed further: https://www.surveymonkey.com/r/RPTKD9L
-- Cheers, Ana
If you have completed reading the first paper (or decided not to complete it at all), plese take a survey (it's very important for future activities): https://www.surveymonkey.com/r/RPTKD9L
The quiz is ready and the link to it will be published here and in #general channel tomorrow. The goal of the quiz is to enhance the paper reading experience. It's not a competition. :) The quiz will be tested by two of our members before I publish it (Marce and Dharmendra). Also, for those who will make it to the implementation stage, Dharmendra will create a quiz that covers practical aspectes of PCA. I saw a few questions - they are great!
The deadline for the reading is tomorrow and I hope most of you will have completed it by that time. But if not - it's still okay (it's soft deadline). There are two more activities we can perform in addition to discussion, which are:
- Paper summary. It would be good if somone can write a summary.
- Poll-quiz on surveymonkey that will help us to understand what was the hardest part of the paper.
PS. Sorry for not updating for a few days. My life has just changed completely. For good. I was too happy. :)
-- Cheers, Ana
Marce (Slack: Marce) suggested that we use ReviewNB in our implementations. In short: it allows to review changes (diff) in notebooks in a human-readable way (not JSON diffs!), which is quite cool. I have already installed this plugin for our organisation, and Marce will come up with more info/tutorials soon! :)
-- Cheers, Ana
We have created an organisation for the club - https://github.com/papersdclub! This will contain implementations as well as organisational notes, reading lists and other useful things. If you haven't done it already - request to join. Alternatively, send me a message on Slack or add you github in the GoogleDoc with members! More updates will follow in the nearest future! :)
-- Cheers, Ana
We've asked for the estimates for the time needed to complete reading. Most of of the club members are confident they will be able to complete reading in one week. Some respondents estimated their reading time to be 2 weeks. Don't worry, guys, we'll wait for you! :)
-- Cheers, Ana
We have selected the first paper for reading. And the winner is "A Tutorial on Principal Component Analysis". The paper was selected after running a survey. It turned out that most people were interested in introductory papers. And, actually, there were two winners: the current paper and word2vec. The next survey will include the newly proposed papers.
-- Cheers, Ana