Song Lyric Toxicity, Commit Assistant, NLP Progress, DensePose, PyTorch Geometric,…
Welcome to the 18th Issue of the NLP Newsletter! Here is this week’s notable NLP news! Today we have a lot of open datasets, computational health studies, bias and ethics for AI, transfer learning for both computer vision and NLP tasks, and much more.
On People and Society…
On safety standards for AI systems and factors to consider (opinion piece) — Link
You think it’s possible to learn accurate cognitive models with no data and little to no human knowledge engineering? If you are curious you will find this work interesting — Link
Christopher Manning talks about neural network that can reason (Video) — Link
Very nice repository to check the current progress of NLP related tasks and research — Link
Learn about the latest work on interpretable machine learning research — Link
Deep Learning for NLP: PyTorch vs Tensorflow (Video) — Link
On Education and Research…
ICYMI, here is the latest work on transfer learning for NLP by OpenAI — Link
Another paper on transfer learning about how to uncover the cluster and relationships (disentangle) between computer vision tasks — Link
A novel spatio-textual clustering algorithm for clustering tweets using heterogeneous data — Link
Detecting social network mental disorders, such as internet overload and net compulsion, through tensor decomposition — Link
On how to learn joint multimodal embedding space with unpaired text and video data — Link
Results are in for the Retro Contest hosted by OpenAI — Link
Learn about how a bidirectional-asynchronous framework is leveraged to generate meaningful emotional replies in conversations (something traditional language models fail to achieve due to generic replies) — Link
On Code and Data…
Here’s a cool website where you can search for machine learning papers that have open source code — Link
A song lyric toxicity dataset is made available accompanied by analysis and slides — Link
An impressive study on how to collect high-quality data through search queries in developing nations, which may have some serious health benefits for society — Link
Code for adversarial training methods for semi-supervised text classification — Link
DeepMind open sources their dataset used to train the generative query networks (GQNs) for neural scene representation and rendering — Link
How to efficient train sequence to sequence models for neural machine translation with Tensorflow (Colab tutorial) — Link
PyTorch Geometric is a tool for implementing geometric deep learning with PyTorch — Link
On Industry…
Here is an AI-based tool that helps make it easier to code video games. The new tool, named Commit Assistant, is offered by Ubisoft — Link
The Facebook research team introduces DensePose, a real-time approach for mapping human pixels from 2D images to 3D surface model of a human body — Link
Understanding emotions for customer support — Link
Detecting sarcasm with deep convolutional neural networks — Link
Learn more about how researchers are making machine learning algorithms fair — Link
Quotes of the day…
Thanks Sebastian for making our lives easier! Deeply appreciate this effort.
Illustrations of the day…
Poster on teaching machines with human language rather than labels — Link
Worthy Mentions…
Sebastian Ruder NLP Newsletter (Issue #26) — Link
Deep Learning for NLP (Lecture slides) — Link
Workshop on deep reinforcement learning (Video presentations) — Link
Learn more about bias and variance in machine learning with very neat visualizations — Link
If you spot any errors or inaccuracies in this newsletter please leave a comment below. Submit a pull request if you would like to add important NLP news here.