Manubot, NALU, GluonCV, ml5.js, doc.ai, OpenAI Benchmark, Deep-Speare, Cyberbullying Detection,…
Welcome to the 24th Issue of the NLP Newsletter! I am Elvis Saravia from Belize. Here is this week’s notable NLP news: OpenAI bots defeat humans in Dota 2 match-up; a cyberbullying detection algorithm; GluonCV helps to build and re-implement computer vision algorithms; debate about meaning and language modeling; an AI lip-reading tool…
On Society…
Anthem is partnering with doc.ai to leverage AI to help predict when people will experience allergies or allergy patterns — link
A new paper discusses an approach to achieve automatic cyberbullying detection in social media (compares with Fasttext) — link
Researchers are trying to develop an AI program that read lips, which can be used to develop a wide variety of helpful applications such as assisting the deaf — link
OpenAI Dota 2 bots achieve remarkable victory (2 to 1) over top human Dota 2 players (the effort is called OpenAI Benchmark) — link
Check out this interesting Twitter NLP debate from leaders in academia discussing the problems of meaning and whether it can be extracted from language models or other deep learning methods — link
On Education and Research…
Researchers develop a machine learning algorithm which uses eye movements to predict personality traits — link
A new paper entitled “An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling” discusses the differences between CNNs and RNNs for sequence modeling (top recommendation) — link
Manubot is an open source tool that enables open online collaborative writing — link
Sebastian Raschka et al. propose a new semi-adversarial networks approach to enhance privacy by generating gender-neutral face images (summary and PyTorch code available) — link
Deep-Speare, a deep learning model trained on William Shakespeare’s sonnets, is able to generate poetry in a similar style to Shakespeare — link
Learn about how an attention-based mechanism is being used to develop multiple instance learning, which is “a variation of supervision learning where a single class label is assigned to a bag of instances.” — link
DeepMind develops an approach to enhance neural networks with Neural Arithmetic Logic Units (NALU) which are able to track time, perform arithmetic over images of numbers, count objects in images, among other capabilities — link
On Code and Data…
ml5.js is an open source web development tool, built on top of Tensorflow.js, that lets users easily access machine learning algorithms and models on the browser — link
Learn about all the different techniques that are easily accessible from the fast.ai library in this fun post — link
The MXNet team releases GluonCV , a deep learning toolkit that lets engineers and researchers quickly develop new algorithms and baselines (for image recognition, object detection, and semantic segmentation) — link
On Industry…
OpenAI develops a human-like robot hand to manipulate physical objects with dexterity — link
Quicksilver is an AI-enabled software tool designed to help Wikipedia editors to fill in the blind spots, such as suggesting missing scientific citations from notable scientist in the the different fields — link
LG opens new AI lab in Toronto and partners with leaders in academia and startup-ups to continue enhancing their products and services — link
Learn how AskArvi — a mobile application — uses NLP and deep learning to understand customer’s needs in order to provide insurance recommendations — link
Quotes of the day…
Worthy Mentions…
Jacob Buckman, Google AI Resident, teaches how to use Tensorflow for beginners — link
Nathan Benaich releases new issue of the nathan.ai newsletter — link
Sebastian Ruder releases new issue of the NLP Newsletter — link
If you spot any errors or inaccuracies in this newsletter please comment below. As usual, if you have any further questions or want to discuss any of the headlines reach me out on Twitter (@omarsar0).
