SWAG Q/A Dataset, PolygonRNN++, Deep Learning in NLP, Diagnosing Retinal Disease, Pix2Pix on the Browser, AutoKeras,…
Hi, I am Elvis from beautiful Belize 😎. Welcome to the 26th Issue of the NLP Newsletter! Here is a quick overview of this week’s notable NLP news: a lot of industry news — from AI-based call centers to improving insurance companies with AI; a pix2pix implementation on the browser using Tensorflow.js; more studying material on deep learning for NLP — from ; a new character-level language modeling technique; new dataset releases; and other important developments in the field of NLP and machine learning.
On People…
Renowned professor and researcher, Pedro Domingos, is set to become the managing director to lead the machine learning efforts at a New York-based investment corporate, D.E. Shaw Group — link
Observe.AI, a startup that uses AI to improve call center services, has raised $8 million to increase efforts to automate the call center experience with machine learning and NLP technologies — link
Vered Shwartz discusses deep learning in NLP in this introductory post, which is aimed for non-experts or NLP enthusiasts 🔝 — link
On Education and Research…
Check out this very nice primer or light introduction to natural language processing (NLP) report by Niklas Donges — link
PolygonRNN++ is a new interactive annotation tool for segmentation datasets. The PyTorch implementation for this system, which was accepted at CVPR 2018, has been released — link
André Martins summarizes his notes and discusses some of the most important topics presented in this year’s Annual Conference of the Association for Computational Linguistics (ACL) and International Conference in Machine Learning (ICML) — link
DeepMind claims to have developed an automated solution for detecting and diagnosing retinal disease using deep learning 🔝— link
A new paper entitled “Character-Level Language Modeling with Deeper Self-Attention” shows how a character-level language model based on a transformer model outperforms RNN variants on two popular benchmarks 🔥— link
Distill releases a new blog post where they reflect on all the challenges they have experienced in their first year operating as a new kind of scientific journal ⭐️ — link
On Code and Data…
A Feedly team discuss their findings on applying transfer learning to a popular classification task using the ULMFit language modeling technique — link
Jonathan Kummerfeld has developed an impressive web page to visualize and showcase a neural Part-of-Speech tagger using various deep learning toolkits such as Dynet, PyTorch, and Tensorflow — link
Zaid Alyafeai releases code for his image-to-image translation system which operates on the browser. The system implements the pix2pix paper and was developed to work on the browser using Tensorflow.js 🔝 — link
DAIR.ai releases new post that provides a guide on how to implement a simple neural network from scratch using PyTorch and Google Colab — link
SWAG is a new natural language inference dataset which will be presented at the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP 2018) conference. The dataset consists of adversarially generated and human verified question/answer pairs, which are useful to test language modeling, question/answer, or natural language inference systems — link
Learn about all the plans for the upcoming Tensorflow 2.0 release — link
Amazon’s new solutions and APIs are set to offer machine learning for offline use, powered by a new compression algorithm, which can help to power in-vehicle infotainment systems and other mobile devices — link
On Industry…
Can AutoKeras, an open-source library that offers neural architecture search functionalities, compete against Google AutoML service? Find out more information in this excellent post by George Seif — link
Learn more about how ML and NLP is being used in the financial sector to redefine services and create new markets — link
On the Dot, a machine learning-based delivery startup, raises £8M to increase their efforts and advance product development — link
Shanghai-based startup, Ins For Renascence (IFR), has raised $18+ million to keep improving their insurance services through data mining and machine learning — link
Google has announced a set of ML developer tools and services (in beta version) for deploying machine learning and NLP model in the cloud — link
Artificial Intelligence is now being used as a tool to level the playing field between small and huge law firms — link
Worthy Mentions…
Learn how to train and serve ML models with tf.keras — link
DAIR.ai releases new tutorial on how to build a recurrent neural network (RNN) with a single layer, consisting of one single neuron, with PyTorch and Google Colab ⭐️ — link
Another great summary of the Neural Arithmetic Logic Units (NALU) paper — link
A new YouTube series on learning how to develop and deploy apps on the browser using Tensorflow.js — link
If you spot any errors or inaccuracies in this newsletter please comment below.
