What's Hot In NLP: How To Participate

Want to join our newest community event? Here's how! This is a translated version of the article. Feel free to report any possible mis-translations in the comments section
Scott Condron
Created on August 26|Last edited on August 26
Comment
Reading papers is all well and good but we believegetting your hands dirty by training models accelerates learning and understanding. That's why we're launching a new W&B community event around growing our understanding of the latest NLP techniques through experimentation!
For this event, we're encouraging you tostart experimenting in publicand sharing what you've found. Why is that? Well, too much machine learning knowledge is hidden in text logs on researchers' remote servers and that doesn't help anyone. Let's come together and grow awareness of the latest techniques that work across different NLP tasks.
For this event, we'll be joined by machine learning engineer Darek Kleczek. He and his team recently came 2nd in a  Kaggle Competitionand we'll be learning from their approach as well as trying to apply it to other NLP problems.
How To Build an Efficient NLP Model
In this article, we look at how we achieved 2nd place on the Kaggle Feedback Prize - Predicting Effective Arguments Efficiency Leaderboard and what we learned. 
﻿
Start ExperimentingAs we did last time around with  Jeremy Howard and the Fast.ai community on Computer Vision models, we want to understand and document how different model architectures and training techniques work on different NLP tasks.  
We're encouraging our community to:
Choose a modern NLP technique to experiment with
Evaluate how it affects the performance on different tasks
Document the technique and your experiments, then share them with the community
Why participate?Shine a light on the best NLP techniques through experimentation.
Grow your skills. Writing things down helps you to organize what you’ve learned, which often lets you uncover gaps in your knowledge.
Meet like-minded people and grow your network. Join a  friendly community of machine learning practitioners, helping one another learn about all things Machine Learning and MLOps.  
We’ll be giving outprizes including GPU credits and W&B merch. We want to encourage high-quality work as much as we can so we’ll reward submissions based on practicality, creativity, clarity, and usefulness for the W&B community.
How to get started:You can develop your own pipeline or use our  starter notebook    
Here's how you submit the W&B Report with your experiments:All submissions should use W&B Reports. You can create Reports from within a W&B Project Workspace by clicking the "Create Report" button on the top right of the page.
To submit, please  share your Report within the W&B Discord.  
⌚TimelineDon’t be a perfectionist. To encourage you to just begin experimenting, the event will run for a period of 2 weeks starting on September 20th. We'll be showcasing the best submissions throughout. We cannot wait to see what you create!
﻿
Add a comment