Pictures to Poetry, Facebook teaches ML, TensorFlow 1.10.0, MnasNet, Bixby Speaker, AI in 2026,…
Hi, I am Elvis from beautiful Belize 😎. Welcome to the 25th Issue of the NLP Newsletter! Here is this week’s notable NLP news: a rare Yoshua Bengio interview on creating a research lab; lots of news related to machine learning, health, and education; exciting new advancements of deep reinforcement learning algorithms; several tutorials on reproducibility and data scarcity; a couple GPU stories and much more.
On People…
Yoshua Bengio shares his experience on creating a research lab and what are the key lessons and best practices researchers should be paying attention to — link
Learn about the challenges of using NLP to assist in improving clinic documentation — link
Professor Toby Walsh has recently published a new fiction book entitled “2062: The World That AI Made” in which he talks about what AI technology will become in the year 2062, when it is approximately expected by many that AI will surpass human intelligence — link
New free e-book reports on the current state of machine learning adoption in the enterprise. The book discusses the different ML strategies and evaluation metrics used by the enterprise — link
Fast.ai students have implemented an algorithm that beats models implemented by Google researchers on an important benchmark — link
A real-world reality check on AI and a discussion about how far AI intelligence is able to go at the moment — link
On Education and Research…
Christopher Manning and Ruslan Salakhutdinov have released a series of overview lectures on deep learning methodologies and how they can be applied to various data settings (highly recommended read) — link
DAIR.ai releases tutorial that teaches how to create a word cloud using Python and Google Colaboratory (Colab). You will learn how to leverage the free storage provided by Google Drive and the powerful, free programming environment of Google Colab — link
DeepMind releases new research that attempts to make sense of the world around us by associating events to sounds, such as the wine glass shattering and the accompanying crash — link
BAIR releases new article discussing the trade-offs between recurrent neural networks and feed-forward neural networks when used to solve various tasks (highly recommended read) — link
On Code and Data…
Anima Anandkumar releases notes and slides on the problem of data scarcity in deep learning and discusses ways to mitigate this — link
ML is coming fast to the cloud. Find out what types of GPU offerings are available in the market in this nicely organized GitHub repository — link
Microsoft researchers release an open source project called TextWorld, which basically attempts to train deep reinforcement learning agents using a series of text-based games — link
The TensorFlow team releases TensorFlow 1.10.0, which includes: BigTable integration, Distribution deprecated, new symbols proposed,… — link
Kaggle is now allegedly offering free access to GPUs on their notebooks — link
On Industry…
Microsoft has trained a conversational agent, known as Xiaolce, to convert pictures into Chinese poetry. This is part of an ambitious initiative to teach machines higher-level natural language generation abilities as compared to simple tasks such as generating caption — link
Learn how Facebook plans to teach machine learning to people in a six-part video series — link
Google release new blog article discussing the machine learning algorithms behind Smart Linkify, an API available on Android 9 that adds clickable links to certain type of entities found in textual conversations — link
Coursera is releasing a new ML-based tool to help employers learn about how and what type of skills their employees are acquiring on the platform — link
A review on Samsung’s smart speaker, known as Bixby Speaker — link
Jeremy Howard, fast.ai cofounder, announces new collaboration with doc.ai, a company launching a data trial to collect data and test whether allergies can be predicted with machine learning algorithms — link
Google researchers have developed a state-of-the-art method called MnasNet, which is an automated neural network architecture search approach for designing mobile ML models using reinforcement learning — link
Worthy Mentions…
COLING 2018 release new blog article which talks about how to improve your presentation skills for academic conferences — link
Vincent Chen, student at Stanford, gives his advice on what math you need for learning machine learning — link
Aurélien Geron releases new video explaining how to reproduce outputs rendered by deep learning models built on TensorFlow. He discusses many important points you should be aware of when reproducing code — link
Learn how you can start using TPUs on Google Colab — link
If you spot any errors or inaccuracies in this newsletter please comment below.
