DeepSuperLearner, Spherical CNNs, Google Semantris, Debater Data, AlterEgo, Text-to-Images GANs, Hate Speech Detection,…
Welcome to the 11th Issue of the NLP Newsletter! Here is this week’s notable NLP news!
On People…
A great roundup of the talk delivered by Rachel Thomas discussing the accessibility in AI at Stanford — Link
Important talk on “Natural Language Understanding: Foundations and State-of-the-Art” — Link
Lila Ibrahim is DeepMind’s new chief operating officer — Link
Emotion Recognition in Context — Link
On Education and Research…
An interesting paper on generating images from scene graphs using graph convolution and GANs — Link
Another interesting research by AI2 on “Scripts to Compositions to Videos” — Link
NLP research discusses the possibility of predicting Twitter user socioeconomic attributes — Link
Paper discusses a novel method for understanding online hate speech — Link
NLP vs NLU: What’s the difference? — Link
On Code and Data…
DeepSuperLearner — an implementation of the deep ensemble methods for classification problems — Link
Dataset release — Recorded Debating Dataset with 60 speeches (audio + ASR & human transcripts) — Link
Building convolutional neural nets (CNNs) in PyTorch — Link
On Industry…
Semantris — word association game powered by NLP and machine learning — Link
Data scientist gives very useful tips on how to build a profile and prepare for the first data science job — Link
AlterEgo transcribe words users “speak silently”; could have major applications in NLP research — Link
Google paper on “Exploring the limits of Language Modeling” — Link
A non-NLP application of word2vec — Link
Quotes of the Day
I had to feature one of these here ;)
Visualizations of the Day
AI Evolution
AlpahGo Zero Cheat Sheet — Link
Worthy Mentions…
Sperichal CNNs — novel method for building convolutional neural networks — Link
Interesting blog post discussing lessons learned in reproducing a deep reinforcement learning paper — Link
Cool blog post discussing ways to make the most out of Jupyter notebooks — Link
