Share this newsletter with friends and colleagues who want to stay up to date with research and news from the Allen Institute for AI.  Subscribe here.

AI2 NEWSLETTER | November 2019

Green AI


The computations required for deep learning research are doubling every few months—these computations are costly, and they have a surprisingly large carbon footprint. A new position paper from AI2 advocates for making efficiency an important criterion for AI research, alongside accuracy and related metrics.
We propose reporting the financial cost or "price tag" of building, training, and running models to help inspire research into more efficient methods. Our goal is to make AI both greener and more inclusive—enabling any inspired undergraduate with a laptop to write high-quality research papers. Learn more:

Semantic Scholar now covers all of science


Semantic Scholar now includes over 175 million academic papers from all fields of scientific study. "Every scientist now has a powerful, free AI search engine at their fingertips," says Doug Raymond, General Manager of the Semantic Scholar team.
Learn more about this important milestone for our free, AI-powered research tool:

AI2 at EMNLP 2019


AI2 had a strong showing at this year's EMNLP conference—special congratulations to the team that won the Best Demo Paper Award for AllenNLP Interpret: A Framework for Explaining Predictions of NLP Models!

NLP Highlights Podcast

Have you heard? NLP Highlights is a podcast from AI2 researchers that invites a variety of researchers to discuss their work in natural language processing. Check out our behind-the-scenes interview with hosts Matt Gardner, Pradeep Dasigi (research scientists at AI2), and Waleed Ammar (research scientist at Google) about their inspiration, advice, and what's next for NLP Highlights.
 

The hosts of NLP Highlights: Matt Gardner, Pradeep Dasigi, and Waleed Ammar.
NEW LEADERBOARDS

DROP

DROP is a QA dataset that tests the comprehensive understanding of paragraphs. In this crowdsourced, adversarially-created, 96k question-answering benchmark, a system must resolve multiple references in a question, map them onto a paragraph, and perform discrete operations over them (such as addition, counting, or sorting). 
 

QASC

QASC is a question-answering dataset with a focus on sentence composition. It consists of 9,980 8-way multiple-choice questions about grade school science, and comes with a corpus of 17M sentences. Each question is annotated with two facts from the corpus that can be combined together to arrive at the answer.
 

MC-TACO

MC-TACO is a dataset of 13k question-answer pairs that require temporal commonsense comprehension. The dataset contains five temporal properties, (1) duration, (2) temporal ordering, (3) typical time, (4) frequency, and (5) stationarity (whether a state is maintained for a very long time or indefinitely).
 
Check out all of our leaderboards at leaderboard.allenai.org →
 
MORE FROM AI2

AI2 has a new logo!

Five years in, it was time to refresh our look with a more streamlined logo and brighter, bolder colors. An updated website is in the works – in the meantime, you can find the various versions of our new logo here

The AI2 Incubator is a key part of the incredible startup ecosystem coming together in Seattle – "We have all of the ingredients and all of the components are in place. So let’s do it."
Learn more: Analysis: Seattle startup ecosystem poised for unprecedented acceleration of company creation (GeekWire)

 

CI with GitHub Actions for Research Code: AI2's Mark Neumann explains how (and why) to set up free continuous integration for your research project using GitHub Actions.


 
AI2 is growing! Check out the Semantic Scholar team's colorful new office space.
 
Were you forwarded this newsletter from a friend or colleague? To stay up to date with research and news from AI2, subscribe here.
Twitter
LinkedIn
Instagram
Website
Copyright © 2019 The Allen Institute for Artificial Intelligence, All rights reserved.

AI2 Newsletter Archive