NLP News Cypher | 01.26.20

Author(s): Ricky Costa

Originally published on Towards AI.

Weekly Newsletter on Natural Language Processing (NLP) News and Research

NLP News Cypher U+007C 01.26.20

Big Bad NLP Database Update

And…. we’re back! How was your week??

We’re continuing our hard work on the NLP Database! During the mid-week, we added 21 datasets and user contribution has helped tremendously, they are mentioned here:

Since the release of the Big Bad NLP Database, we have received overwhelming support across the globe including visitors stemming from 98 countries in the past week alone. We have many more datasets left to go and hopefully, we’ll capture more international variants. More updates coming soon!

This Week:

The Dutch RoBERTa

Hacking GitHub for Blog Posting

Serverless, VMs and Containers

Mellon’s Twitter NLP Library

A New Dataset for Visual Question Answering

AI Content Made Simple

mmmmmBART

Know Your Hardware

Dataset of the Week: DailyDialog

The Dutch RoBERTa

Looks like we have a new language-focused transformer, and it’s straight out of the Netherlands, world meet RobBERT. It achieves SOTA results on several Dutch-based downstream tasks. You can read more about it here:

Blog: https://people.cs.kuleuven.be/~pieter.delobelle/robbert/

GitHub (silhouette head shot, props):

iPieter/RobBERT

A Dutch language model based on RoBERTa with some tasks specific to Dutch. Read more on our blog post or on the paper…

github.com

Paper:

Hacking GitHub for Blog Posting

A fellow figured out how to embed interactive Jupyter notebooks in GitHub pages U+1F60EU+1F60E. Pretty cool if you want to show your work with a little more twinkle.

Way to export notebook as HTML in Jekyll for Blog Posts?

The problem I'm trying to solve is writing blog posts with Jupyter, but keep my Altair charts interactive. Furthermore…

forums.fast.ai

Serverless, EC2, and Containers

Your next steps to deploying your AI model requires a unique environment. In this white paper, Flux7 details the differences between three possible cloud frameworks:

“Serverless is an execution model where the cloud provider is responsible for executing a piece of code by dynamically allocating the resources. Serverless resources are priced by the amount of resources consumed; you pay only for what you use to run the code.”

“A container is a standard unit of software that packages up code and all its dependencies so the application runs quickly and reliably from one computing environment to another”

EC2: Amazon’s VM lol

Paper:

Mellon’s Twitter NLP Library

Hey… want to checkout Carnegie Mellon’s Twitter NLP library? You get a tokenizer, a part-of-speech tagger, hierarchical word clusters, and a dependency parser (and more!). It’s in Java but in the blog, they provide links to Python wrappers.

Twitter Natural Language Processing — Noah's ARK

We provide a tokenizer, a part-of-speech tagger, hierarchical word clusters, and a dependency parser for tweets, along…

www.cs.cmu.edu

A New Dataset for Visual Question Answering

“Sub-VQA, a new dataset consisting of 200K new perception questions which serve as sub questions corresponding to the set of perceptual tasks needed to effectively answer the complex reasoning questions in the Reasoning split.”

Ok, so what the hell does this mean?

Here’s an example from the paper: the model sees a picture of a banana. The model is asked, “Is the banana ripe enough to eat?” It correctly answers “yes”, but then the model is asked “Are the bananas mostly green or yellow?” and it incorrectly answers green showing an inconsistency of logic from the 1st answer. This new dataset attempts to look into this issue.

Paper:

AI Content Made Simple

If you want to see how the AI community collates content, you should check this platform out:

Top Content – practicalAI

Discover and organize the top community-curated content.

practicalai.me

They make it simple to add tags to URLs so you stay on top of content coming in every week. This is essentially a clever content management platform for top AI content! Great Job!

mmmmmBART

Facebook throwing heat with a NEW Seq2Seq multi-lingual model for machine translation called mBART. We found one of the researchers on Twitter. Here is the thread discussing the research in detail:

A thread written by @thoma_gu

1/7] Super excited to present our recent work — mBART.We demonstrate multilingual denoising pre-training produces…

threader.app

FYI, GitHub code will soon be released on Fairseq’s repo according to the researcher.

pytorch/fairseq

Fairseq(-py) is a sequence modeling toolkit that allows researchers and developers to train custom models for…

github.com

Paper:

Know Your Hardware

In a previous newsletter, I raved about Vivienne Sze’s workshop at NeurIPS and she blew me away. Partly because she had so much knowledge to share and partly because I understood the pain of deploying large models while dealing with compute/memory/inference issues. If deployment pain interests you, check out her lecture on Fridman’s podcast:

Dataset of the Week: DailyDialog

What is it:

A multi-turn dialog dataset that is labeled with communication intention and emotion information.

Sample:

Where is it?

Yanran's Attic

We develop a high-quality multi-turn dialog dataset, DailyDialog, which is intriguing in several aspects. The language…

yanran.li

Every Sunday we do a weekly round-up of NLP news and code drops from researchers around the world.

If you enjoyed this article, help us out and share with friends or social media!

For complete coverage, follow our twitter: @Quantum_Stat

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI