Facebook’s SapFix, Improvisational Comedy, Textual Analogy Parsing, Ethical ML, Sound Search, GANs and Art,…
Hi and welcome to the 30th Issue of the NLP Newsletter! I am Elvis from Belize, Editor of dair.ai, and a PhD researcher in AI and NLP. Here is this week’s notable NLP news: ethical machine learning; lots of NLP tutorials; GANs for art generation; NLP-based multitask learning; Reinforcement learning in multitask environments, and much more.
On People…
Last week I reported about a new hate speech paper. This week there is an article discussing how these type of online hate speech detectors can easily be tricked by humans — link
In a candid talk delivered at the EmTech conference, AI researcher, Zachary Lipton discusses why he believes that AI has been over-hyped and how to promote a much healthier conversation around AI — link
Learn more about abstraction in neural network in this fun talk delivered at Eyeo Festival by David Ha — link
Listen to Andrew Ng, a popular influencer in AI, talk about his journey from early days in Google Brain to the recent deeplearning.ai course (podcast) — link
Vered Shwartz releases a new blog post where she discusses the importance of ethical machine learning and what questions we should be asking when building ML systems that make important decisions — link
Learn how NLP and machine learning are being used to train bots to create dialogue and perform improvisational comedy — link
On Education and Research…
Richard Socher keynote talk on the recent progress in NLP and areas such as multitask learning — link
A very light introduction to neural attention and how they are used for image segmentation and neural machine translation — link
Helena Sarin published an article in The Gradient explaining how she uses GANs to generate artistic pieces — link
Researchers at MIT develop an AI system that is able to achieve human-like reasoning to evaluate images. The system relies on an “attention mask” which enables a visual understanding of how the model reasons in several sub tasks — link
New paper proposes a system called Textual Analogy Parsing (TAP) to model higher-order meaning based on a frame-style meaning representation. For instance, the technique is able to specify what is shared and what is compared between component facts — link
DeepMind develops a new technique called PopArt which enables a reinforcement learning agent to perform accurately in multitask environments — link
Grammarly, a popular writing tool, set to gradually release a beta version of Chrome browser plugin that works with Google Docs. The company claims that they are leveraging NLP and AI to help people improve their writing — link
On Code and Data…
Listen to Michael Nielsen talk about how important Open Access is for achieving impact and his experience in reaching 3.5 million readers with his book “Neural Networks and Deep Learning” — link
Code walk-through on variational autoencoders with PyTorch — link
SemEval 2019 Task 6 involves an interesting dataset for categorizing and identifying offensive language in social media — link
Great tutorial on how to get started with the AllenNLP library to build and train your NLP models — link
On Industry…
Microsoft’s latest patent intends to the take the phone caller ID feature to the next level using NLP. The system will be able to receive a call and identify caller intention and action — link
Apple announced three iPhones with an A12 chip containing a neural engine, which is able to perform 5 trillion operations per second — link
SapFix is an AI programming tool developed by Facebook which is able to scan code for bugs and suggests patches — link
Microsoft acquire Lobe, a machine learning startup that offers drag and drop machine learning tools similar to Azure ML Studio — link
Google releases a new version of Sound Search, which allows an android phone to recognize and search for music — link
Worthy Mentions…
Find out how Boeing, the airplane company, is investing in AI to help pilots — link
Want to get started with reinforcement learning, check out this nice Medium article that provides a comprehensive introduction to reinforcement learning — link
If you spot any errors or inaccuracies in this newsletter please comment below.