Libratus, OpenAI Five, Common Sense AI, ARNN, Positive AI, WaveNet Stack, DARTS, ROAR,…
Welcome to the 19th Issue of the NLP Newsletter! Here is this week’s notable NLP news! Lots of reinforcement learning news, adversarial reprogramming of neural networks, introduction to sound classification, interpretability methods for deep learning, Microsoft and Amazon new dataset releases, $125 million initiative to solve common sense AI, and much more.
On People…
Learn more about Libratus, the world’t best AI-based poker player (overview by The Gradient) — Link
It is time to stop worrying and love AI instead? — Link
University of Washington professor, Yejin Choinka, to lead $125 million “common sense AI” initiative — Link
Famous language expert, Jeff Elman, dies at 70 — Link
Sam Altman discusses progress in reinforcement learning — Link
Can AI beat human doctors on a sophisticated clinical exam? — Link
CVPR makes public a set of educational material for learning how to properly conduct research and be involved in the community (videos and slides available) — Link
Why scientists are encouraged to have a social media presence — Link
On Education and Research…
A great introduction to how sound classification is done — Link
An introduction on how to use text and NLP techniques to conduct economic research — Link
New benchmark results for the Stanford Tokenizer — Link
Capturing long-range structures in music by stacking WaveNet autoencoders (listen to samples generated here) — Link
Combining domain knowledge and deep learning for conducting sentiment classification — Link
Forcing neural networks to behave as if they were trained on a completely different task (Adversarial Reprogramming of Neural Networks (ARNN)) — Link
Face generation from text — Link
On Code and Data…
DARTS: Differentiable architecture search — an algorithm to efficiently design high-performance convolutional architectures for image classification — Link
On reproducing “World Models” with PyTorch — Link
How deep learning with synthetic data will democratize the tech industry — Link
Amazon releases sales rank dataset for kindle and print books — Link
Remove and Retrain, ROAR, a benchmark that can be used to evaluate the accuracy of interpretability methods of deep learning networks — Link
Microsoft has just released 125 million building footprints in the US as open data — Link
On Industry…
OpenAI Five is the agent trained on reinforcement learning that’s defeating amateur human teams on the complex game of Dota 2 — Link
Disney research reveals a method that automates a learning environment for developing control policies directly on the hardware of a modular legged robot — Link
Quotes of the day…
Illustrations of the day…
Explaining backpropagation through a storytelling approach — Link
Worthy Mentions…
All the things you wish you knew before starting to use Tensorflow — Link
Learn more about transformers and how they are used to achieve state of the art results for many NLP tasks — Link
Keras vs PyTorch: Where to Start? — Link
Leveraging knowledge graphs and deep learning for video classification and recommendation — Link
If you spot any errors or inaccuracies in this newsletter please comment below. Submit a pull request if you would like to add important NLP or AI news here.
