NLP News Cypher | 04.05.20

Author(s): Ricky Costa

Originally published on Towards AI.

NATURAL LANGUAGE PROCESSING (NLP) WEEKLY NEWSLETTER

NLP News Cypher U+007C 04.05.20

Meditations

When Lex isn’t interviewing AI pioneers in a sullen dark suit, he takes to the Twitterverse to question the physics of intercourse in the vacuum of space.

I got 1 like for this tweet, it was from HAL 9000.

How was your week? U+1F60E

We’ve updated the Big Bad NLP Database. We added 38 new datasets, taking the total past 350! Thank you to Tommaso Pasini, Henry Dashwood, Bill Lin, Reid Pryzant , Parth Parikh and Christian Hardmeier for your contribution!

Have a dataset to share? Then, “Break on through to the other side” (a.k.a. please share the dataset by clicking on the “Contact” link on the BBND webpage and sending over the details.)

BTW, we have a surprise for you next week. Stay Tuned! U+1F9D0

This Week:

Visual Storytelling

Wolfman Cometh

Matplotlib Ready for Prime Time

Decomposition Redux

Stanza’s Notebooks

ML Inference on the Edge

Reinforcing NLP

Dataset of the Week: ALFRED

Visual Storytelling

A new model was released recently discussing the topic of visual storytelling via reinforcement learning! What’s visual storytelling?

“given a photo stream, the machine is trained to generate a coherent story in natural language to describe the photos.”

The model, ReCo-RL, is rewarded on “relevant” storytelling based on 3 NEW criteria: relevance, coherence and expressiveness. What’s cool is that this model achieves excellent performance on traditional and the new criteria.

GitHub:

JunjieHu/ReCo-RL

Implemented by Junjie Hu Contact: [email protected] If you use the codes in this repo, please cite our AAAI2020 paper…

github.com

Paper:

Wolfman Cometh

The T5, that big Google model, that’s both an encoder and decoder can now be accessed via the Transformer library. If you want to take the T5 for a test drive with summarization or translation, check out my Colab notebook below based on U+1F917’s original notebook.

Colab:

Google Colaboratory

Edit description

colab.research.google.com

Matplotlib Ready for Prime Time

One of my biggest pet peeves with Matplotlib’s graph visualizations was that it looked like it was running on windows 95. But recently, I found this blog that shows how you can impress your data science friends with some impressive visualizations on Matplotlib. Includes code!

Matplotlib Cyberpunk Style

Let's make up some numbers, put them in a Pandas dataframe and plot them: import pandas as pd import matplotlib.pyplot…

matplotlib.org

Decomposition Redux

I had previously discussed this model in a previous newsletter. But this time they have returned with a Github repo! Won’t dive in again, but in essence, this model decomposes multi-hop questions into simpler ones to assist in question answering:

GitHub:

Unsupervised Question Decomposition for Question Answering

We improve automatic question answering (QA) by decomposing hard questions into easier subquestions that existing QA…

medium.com

Blog:

Unsupervised Question Decomposition for Question Answering

We improve automatic question answering (QA) by decomposing hard questions into easier subquestions that existing QA…

medium.com

Stanza’s Notebooks

You may have heard of Stanford’s Stanza: a new multi-lingual NLP Python library.

What you may have not of known, is that they have awesome colabs to get you jump-started.

Beginner’s Guide:

Google Colaboratory

Edit description

colab.research.google.com

CoreNLP Guide:

Google Colaboratory

Edit description

colab.research.google.com

ML Inference on the Edge

From the TensorFlow blog: a new delegate release for those dreaming of one day deploying ML models to mobile devices U+1F601.

Today, we are excited to announce a new TensorFlow Lite delegate that uses Apple’s Core ML API to run floating-point models faster on iPhones and iPads with the Neural Engine. We are able to see performance gains up to 14x (see details below) for models like MobileNet and Inception V3.

Blog:

TensorFlow Lite Core ML delegate enables faster inference on iPhones and iPads

April 02, 2020 – Posted by Tei Jeong and Karim Nosseir, Software Engineers TensorFlow Lite offers options to delegate…

blog.tensorflow.org

Reinforcing NLP

Want to get up-to-date on Reinforcement Learning in NLP? We found a repo to keep you occupied during your in-home stay.

jiyfeng/rl4nlp

Reinforcement learning for natural language processing reading group – jiyfeng/rl4nlp

github.com

Dataset of the Week: ALFRED

What is it?

“ALFRED dataset contains 8k+ expert demonstrations with 3 or more language annotations each. It’s a benchmark for learning a mapping from natural language instructions and egocentric vision to sequences of actions for household tasks.”

Sample:

Demo

Controls To enable or disable the game's controls. Look around with your mouse. / Keys to move. Left Click to interact…

ai2thor.allenai.org

Where is it?

askforalfred/alfred

The ALFRED dataset contains 8k+ expert demostrations with 3 or more language annotations each. A trajectory consists of…

github.com

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!

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