Leaderboard

(Updated daily)

RankName(s)TitleAvg. Expert ScorePolular ScoresDirect Challenge ScoreOverall Score
# of ViewsAvg. Watch Time# of Likes# of CommentsOverall Popular Score
1MK2.6614710.1962521517.2309.9
2JL3.0916100.174181444.2507.3
3PS3.647450.3482363.3207
4GH, LP3.813330.3054250.6904.5
5IR3.044580.2934851.2404.3
6AC3.691480.26370003.7
7AL3.631600.29202003.6
8RC32520.2891410.2103.2
9VM3.071120.25940003.1
10SG2.842800.233700.0902.9

AISC 5 min paper challenge

Update: the deadline for video submission has been extended to April 29.

(Please send your inquiries and paper title to [email protected])

If you understand a paper well, you should be able to explain it in 5 mins. Show us if you have what it takes!

Objective

AISC has been creating a video repository of recent and foundational machine learning papers through the in person sessions we host. To complement these, the objective of this challenge is to create a library of bite sized videos that quickly introduce interesting and relevant topics, and allow the audience to quickly get the gist of the recent contributions.

Challenge

In this challenge, we ask you to tell a story that communicates the interesting parts of your favorite recent machine learning paper. The videos will be judged in phase 1 based on their technical and production quality, and in phase 2 based on the engagement they create on YouTube.

  1. Select your favorite machine learning paper published since 2009
  2. Send us an email and tell us what paper you will be working on ([email protected]); we will also add you to our slack channel so that you can brainstorm with everyone
  3. Create a 5 min video about the paper
  4. Put it up on some cloud storage (eg dropbox) and send us the link: submit link below
  5. If your video is selected for phase 2 (see below), advertise it and make sure you get engagement
  6. Win a cash prize

Please note that you can only resubmit your video twice, and submission can be done on behalf of a team

Submit

Important Dates

Submission OpensMarch 29, 2019-
Submission ClosesApril 15, 2019 at 11:59 pm ESTApril 29, 2019 at 11:59 pm EST
Phase 1 qualification resultsApril 15 - April 22, 2019April 29 - May 6, 2019
Publication on YouTubeApril 22, 2019May 6, 2019
Winner announcementMay 6, 2019May 20, 2019

Prizes

1st Place$400
2nd Place$100
all videos qualified for phase 2$10 gift cards

Video Requirements

A qualifying entry should:

  • Tell a 5 minute long, coherent story that incorporates all the important content of the paper; there is no minimum length; longer videos will be clipped at 5-min mark
  • Have a title that is representative and interesting
  • Must be in English

Videos should include:

  • ~1 min, intro and context
  • ~2 min, concept and key takeaways
  • ~1:30 min, results and discussions
  • ~30 sec, wrap up and your verdict

The target audience

You can assume that the audience of your video has an overall knowledge of machine learning, but they have not read the paper you are talking about or ones similar to it.

Eligibility to participate

There are no minimum requirements for participation.

Phase 1 Qualification

Videos in phase 1 will be selected based on the following criteria:

  • Quality of Science Communication [1-5, 5 best]
  • Originality and Creativity [1-5, 5 best]
  • Quality of Production (audio and video) [1-5, 5 best]
  • Technical Soundness [1-5, 5 best]
  • Uniqueness (if more than one video for the same paper is submitted, the best 1-3 videos would be selected for phase 2)

Phase 1 Score: Average of the 4 quantative scores above judged by an expert panel

The minimum score required for qualifying is 2.5

Phase 2 assessment

Award winners will be announced based on the overall engagement generated on YouTube. In order to give all videos equal opportunities, they will all be published on AISC channel at the same time, and advertised through AISC public channels. Advertisement and general awareness about individual videos is up to the participants and highly encouraged.

Phase 2 score: 1/60 x softplus(# Views - 60) + 1/20 x softplus(# Likes - 20) + 1/10 x softplus(# Comments - 10)

Note: Number of views, likes, and comments are multiplied by the "average watch percentage" for each video to better represent the engagement created by the video

Directly challenging others

For every person that you publicly challenge to participate (don't forget to use #AISC_5min), and if they submit a qualifying video, you would get 2 point

direct_challenge_score: 2x # people that mention in their submission that you challenged them

Overall Score

phase_1_score + phase_2_score + direct_challenge_score

In case of a tie in the overall score, it will be broken by the phase 1 score first, and then phase 2 score.

Production quality tips

  • Select interesting papers
  • Try to draft your thoughts first and practice a few times before recording to make sure you have a coherent story; maybe run it by a friend first? And then share the prize if you win with them?
  • Providing enough context is important in helping people understand and engage with your content; spend some time understanding the work around the current one to be able to place in the right context
  • Make sure your environment is quiet
  • Use a good mic (at least a headset)
  • Use a well lit room with clean and non-distracting background
  • Try editing your video if there are parts that you want to omit
  • Do some research on how to present ideas in a concise and engaging way (YouTube has some good content on this)
  • Sound excited and try to use verbal and body language that invites your audience to participate
  • Cat videos are very popular, maybe explain the paper to your pet or plant? Just joking… or am I? Well, as long as what you say is legit and technically sound, why not? Just make sure it doesn’t end up being just fluff!

How to make a video?

You can do this in a few different ways

  • just make a video of yourself talk. perhaps the easiest and fastest option, allows you to get up and running faster, and iterate more easily. Using your phone or computer laptop is the easiest option for this
  • just make a video of your screen as you go through some slides. perhaps the best option for those who prefer to put more focus on the slides. There are lots of screen recording tools that you can use. Just pick your favorite.
  • show both yourself and your slides. This option is more complex and needs more set up but would look nicer. You can use tools like Online Broadcasting Studio to do that.

If you want to participate and uncertain about how to make your video, drop us a line and we will try out best to help

[New] Tips on video editing & scripting (April 23)

Below is a summary of our remote office hour on April 23, where we discussed tips and exchanged ideas on video production.

  • What video editing software to use? Lightroom, Blender, iMovie are all good choices. Some participant also suggested Adobe Presenter.
  • I recommend recording continuously, if you make a mistake, redo the part repeatedly until it's good; later use one of the video editing tools to cut out all the mistakes
  • It's easier to do audio and video at the same time, so that you won't need to align the two tracks later
  • Scripting is helpful in planning things out and budget time exactly the way you want
  • Try to simplify and shorten sentences both spoken and on screen. This was challenging but also rewarding because it forced us to be as simple and clear as possible
  • How to improve audio quality? Use a better mic; alternatively, record voice on your phone while recording your screen at the same time
  • Starting early is important.
  • Rehearse in front of your friends. It can help a lot.
  • We have a hard limit of 5 minutes. Under such short time frame, empirically number of slides should be somewhere between 8 and 15

FAQ

can we extend the deadline? yes, April 29th it is

would i get points for challenging others on social media to participate? yes, for every person that you publicly challenge (#AISC_5min) and if they submit a qualifying video, you would get 2 point

can you share sample videos? yes, soon

can we create teams instead of going solo? of course, even encouraged

is including the code necessary? if it helps with clarity of the story you are telling

how technical does it need it to be? it must be about algorithms or methodologies or applications, and shouldn't be too high level compared to the typical scientific journal standards

can we use graphics/ visualizations? that might actually help people understand the concept better

is resubmission allowed? yes, within limits

would you provide technical feedback on the ML and the video production? we don't promise but if you submit early enough then we might have enough time to do that

can AISC provide recording equipment? since there are participants that are not in Toronto, it would not be fair

We can’t wait to see your videos!

(Please send your inquiries and paper title to [email protected])

List of submitted papers

Date of EntryTitle
2019-04-03EIE: Efficient Inference Engine on Compressed Deep Neural Network
2019-04-03Training a Binary Classifier with the Quantum Adiabatic Algorithm
2019-04-03Session-based Item Recommendation in E-Commerce on Short-Term Intents, Reminders, Trends, and Discounts
2019-04-03Attentive Neural Processes
2019-04-03Image Classification at Supercomputer Scale
2019-04-04On the Classification of SSVEP-Based Dry-EEG Signals via Convolutional Neural Networks
2019-04-07Image-to-Image Translation with Conditional Adversarial Networks
2019-04-10Using convolutional neural networks to explore the microbiome
2019-04-10Combining satellite imagery and machine learning to predict poverty
2019-04-11Posterior Attention Models for Sequence to Sequence Learning
2019-04-12Denoising Time Series Data Using Asymmetric Generative Adversarial Networks
2019-04-14A Style-Based Generator Architecture for Generative Adversarial Networks
2019-04-15Folding: Why Good Models Sometimes Make Spurious Recommendations
2019-04-18Clustering with Deep Learning: Taxonomy and New Methods
2019-04-19Classification of sentiment reviews using n-gram machine learning approach

5 Mininute Paper Challenge is generously sponsored by