ELMo, GLoMo, FloydHub Workspaces, AI Principles, NCRF++, TorchFold, AI Talent Report,…
Welcome to the 17th Issue of the NLP Newsletter! Here is this week’s notable NLP news: Deep contextualized representations (ELMo), new IDE for deep learning, AI Talent Report 2018, Google’s AI principles, graph-based representations for transfer learning, boosting Python-based, NLP modules, and much more…
On People…
Global AI talent report for 2018 describes how the AI talent pool is distributed across the world — Link
Great podcast episode by Microsoft Research on how to make simple models more accurate and accurate models more intelligible or interpretable — Link
NAACL’s best paper award goes to ELMo (deep contextualized word representations) — Link
Very nice talks about RNNs and beyond — Link
On Education and Research…
OpenAI proposes a transformer-based language model that is useful for a wide variety of NLP tasks (inspired by ELMo and CoVE) — Link
A list of some of the most influential papers in deep learning (summaries included) — Link
Training 10,000-layer vanilla CNNs (Paper) — Link
Transcribing music through reinforcement learning — Link
Learn more about why batch normalization works (Paper) — Link
Analyzing behavior of visual questions answering models to identify strengths and weakness — Link
On Code and Data…
TorchFold a tool for PyTorch that makes it easy to batch anything regardless of the complexity of your dynamic architectures — Link
NCRF++, an open-source neural sequence labeling toolkit — Link
HuggingFace introduces NeuralCoref — coreference resolution done via neural networks and SpaCy — Link
Here is a nice dataset which contains short jokes scraped from various websites — Link
Learn how to speed up your Python NLP modules by 50–100 times faster — Link
On Industry…
Google proposes its new AI principles and practices — Link
Introducing FloydHub Workspaces — a new cloud IDE for deep learning — Link
Leveraging latent relational graph-based representation (GLoMo) for enabling transfer learning to many NLP tasks (by Facebook AI Research) — Link
A reinforcement learning environment for self-driving cars built on the browser using Tensorflow.js — Link
Worthy Mentions…
Slides by Adrej Karpathy, on building the software 2.0 stack and what a machine learning IDE should contain — Link
A comprehensive review of deep learning for objection detection — Link
A nice summary of ULMFiT, a transfer learning methods that can be applied to several NLP tasks — Link
If you spot any errors or inaccuracies in this newsletter please comment below.
