lIRO NLP Hackathon
5 November 2022 / Politehnica University of Bucharest
#machinelearning
#nlp
#romanian
#hackathon
#mentoring_sessions
#UPB
#talk
#transformers
Come join the LiRo Machine-Learning Romanian NLP Hackathon
Meet LiRo - a benchmark for Machine Learning NLP models for Romanian! See if you can do better than the current state-of-the-art on selected tasks like Named Entity Recognition or Emotion Detection from Romanian tweets. Try your hand at Diacritic Restoration with the help of the Romanian WordNet and the latest transformer models!
Hesitating that you might not have the full technical knowledge to participate? No worries, we have a workshop session in the morning where we'll dissect a model for each task - you'll learn how to build models that you can further compete with in the hackathon!
Come join Sebastian's keynote, it's open for everybody on Google Meet: https://meet.google.com/vhe-wohw-uky
NEWS:
17.10.22: Full colab notebooks with technical details for the 1st challenge are available!
18.10.22: Sebastian Ruder, Google NLP researcher, will hold a virtual keynote talk about 'Challenges and Opportunities in NLP for Under-represented Languages', come check it out!
26.10.22: Agenda updated
02.11.22: Colab notebook for the Diac Restoration Challenge available.
Prizes
There's a almost 2000 EUR in prizes in two challenge tracks. See updated FAQ for eligibility.
Who can attend?
Everybody, either on your own or in a team.
WHEN ?
Saturday, 5 November 2022,
from 9 AM to 9 PM
WHERE ?
'Politehnica' University of Bucharest, Automatic Control and Computers (ACS), EC105
Register here until the 2nd of November 2022 - registration is free, and takes 5 minutes
cHALLENGES
You can participate in one or both of the following challenges. The LiRo Challenge has 4 tasks, feel free to try your hand at any number of them.
LiRO NLP TASKs Challenge
Pick one or more of the 4 LiRo benchmark tasks:
Prizes are 200E / 100E per task - totaling up to 800E / 400E
Diacritic RestOration
Develop an ML model that "restores" Romanian text by correctly filling in diacritics.
Prizes 500E first place, and 250E second place
Judging Criteria
Project will be judged by performance, complexity and innovation.
Each challenge has a predefined test set (visible or hidden) and appropriate metrics which the participants can use out-of-the-box. However, while performance is important, judging will also take into consideration model size, speed and overall design. If you can achieve 95% performance with only 25% of parameters and some clever use of extra resources, then hats off to you!
Agenda
⫸ 9:00 - 9:15 Hackathon Opening
⫸ 9:15 - 9:45 A word from our Sponsors
⫸ 9:45 - 10:30 Task descriptions
⫸ 10:30 - 11:15 Virtual Keynote with Sebastian Ruder on NLP (don't miss it!)
⫸ 11:15 - 14:30 Mentoring sessions with lunch break (finish time variable)
⫸ 11:15 - 19:00 Coding time! Food and coffee will be available ;)
⫸ 19:00 - 20:00 Project submission and Judging session
⫸ 20:00 - 21:00 Awards and closing session
Coding will take place in ACS's hallway, all the sessions, including opening/closing will take place in EC105.
Sponsors
Organizers
Main organizers:
Stefan Dumitrescu (Adobe)
Traian Rebedea (AIRomania, RoboSelf)
Viorica Patraucean (AIRomania, EEML, DeepMind)
Volunteers:
Mihai Ilie (Sustainalytics)
Andrei Pruteanu (SenseTask)
Alexandra Ciobotaru (DRUID AI)
Mihai Badea (DRUID AI)
Vlad-Constantin Lungu-Stan (Adobe)
Contact
Got questions? Check out the F.A.Q. below or send an email to contact @ airomania.eu .
Don't forget to register here until the 2nd of November.
F.A.Q.
⬜ Who can join?
Everybody. This is an open event for the entire country.
⬜ When should I register?
As soon as possible. The sooner we know how many participants there will be, the better we can organize our logistics (food, drinks, working spaces).
⬜ What is the maximum team size?
Teams are limited to 4 participants.
⬜ Can I join on my own?
Yes, you can go solo.
⬜ Who is eligible for the prizes?
Students. We want to promote ML/NLP to students, so our sponsors' funding goes into prizes and GCP credits for them.
⬜ Is this an online event?
No, this is an on-site hackathon. However, in very special circumstances we might allow a fully on-line team (contact us on email if this is the case). Normally, at least one team-member has to be on-site.
⬜ Can teams participate in more than one challenge?
Yes, but (if eligible) they can only receive a single prize, corresponding to their best result on that challenge. On the LiRo challenge, one team can accumulate all per-task prizes (so you could go up to 800EUR). For example, if you win 1st prize on 3 out of 4 LiRo challenges and also 1st prize on the Diacritic Restoration, you'll automatically receive the larger 600E prize from LiRo vs the 500E you would get from the Diacritic challenge.
⬜ Do I have to bring my own laptop?
Yes, you're expected to have your laptop with you. We'll provide credits for GPUs you can use during the hackathon.
⬜ What do I have to do exactly in the hackathon?
Choose your challenge and obtain the best performance possible; you'll have all the technical details once you register. Don't forget we're also judging by innovation and model efficiency. Winners are expected to present their models in a 5-minute showcase at the end of the hackathon.
⬜ What is the mentoring session?
The morning mentoring session is for those that are hesitant to participate directly in the hackathon. We'll take a challenge step by step and solve it together. To benefit the most, you should have minimal knowledge about how to train a PyTorch model.
⬜ What about food?
We'll provide food, coffee and drinks during the event, no need to bring your own.
⬜ Other questions not covered here?
Please send us an email at contact @ airomania.eu and we'll answer shorty.