Google Duplex, Sense Embeddings, hyperdoc2vec, CoDraw, Keras CPPN, TFLite, Resume Classification…
Welcome to the 14th Issue of the NLP Newsletter! Here is this week’s notable NLP news!
Real-time human pose estimation on the browser (PoseNet to Tensorflow.js)- Link
A talk discussing on how to get started with Tensorflow’s high-level APIs - Link
Using NLP to classify resumes using convolutional neural networks - Link
Towards ML fairness using Generative Adversarial Networks (GANs) - Link
Strategies for optimizing numerical code (slides) - Link
Reading in between sarcastic posts with attention models and deep nets - Link
Learn the basics of creating and operating PyTorch tensor objects - Link
On Education and Research…
Interrogating sentence embeddings about linguistic information (Paper) - Link
Speeding up Tensor contractions with extended BLAS kernels - Link
A very nice comprehensive review of distributed representations - Link
Google releases Google Duplex, a "new technology for conducting natural conversations to carry out “real world” tasks over the phone” - Link
Distributed representations of hypertext documents (hyperdoc2vec) extracted from academic papers for two tasks (i.e., paper classification and citation recommendation) - Link
Summarization for fast.ai new deep learning course - Link
This paper describes the limitations of unsupervised bilingual dictionary induction that rely on x-lingual embedding models, which is an important topic in machine translation - Link
This work investigates whether incorporating subword information in counting models is effective and how important it may be to learn representations of rare and out-of-vocabulary words - Link
An opinion paper discussing strengths and weaknesses of deep nets - Link
On Code and Data...
Learn the internal architecture of PyTorch with this excellent post by Christian S. Perone - Link
Facebook releases dataset, CoDraw, used for training a task that enables collaborative drawing between two agents using natural language understanding from dialog - Link
How AI and NLP algorithms will change healthcare when we start to rely on them for making final decisions - Link
Google io18 event session on TFLite, TensorFlow Extended, TensorFlow Hub & new innovations in TensorFlow Serving - Link
DeepMind’s new paper published in Nature discusses "how grid-like representations in artificial agents enabled shortcuts in virtual environments through vector-based navigation” - Link
Quotes of the day...
Illustrations of the day...
Tensor contractions - Link
Learn how Google is planning to encourage well-being with the use of technology - Link
Learn about how graph neural nets can help to infer latent relational structure and model multi-agent and physical dynamics - Link
Implementation of a Compositional Pattern Producing Network (CPPN) for extreme learned image compression in Keras 2 - Link
If you spot any errors or inaccuracies in this newsletter please open an issue. Submit a pull request if you would like to add important NLP news here.