Music Translation; LDA on Mental Health; Transfer Learning for NLP
Weekly Reading List #6
Issue #6: 2018/05/21 to 2018/05/27
First of all, a big thank you to those who have been supporting this series. It has surpassed my expectation, but unfortunately this series is losing momentum, mostly because of my lack of energy. Therefore I’ve decided to suspend this series after this issue, and instead focus on creating more original and in-depth contents on this platform.
Data Science Tweets of the Day (Shameless Plug)
I’ve created this website www.datasciencetweetsoftheday.ml to keep track of the best data science-related tweets of the day. A lot of it is still WIP, but I’d love to hear your feedback!
Today we have a spell checker based on word vectors, and noising and denoising natural language to get a corpus for…www.datasciencetweetsoftheday.ml
Universal Music Translation Network
This is arguably the most impressive work of the week. Just listen to the samples:
Their model uses a shared WaveNet encoder for all domains, and multiple domain-specific WaveNet decoders. They employs a domain confusion loss to ensure the encoder is not domain-specific.
Investigate the Discourse on Metal Health over Time
This is a very well-executed and well-explained end-to-end NLP analysis project.
For this project I set out to investigate the contexts in which ‘mental health’ has been brought up over time. For this…towardsdatascience.com
Transfer Learning for NLP
Sebastian Ruder just gave a keynote on Successes and Frontiers of Deep Learning:
One of the new frontiers is NLP transfer learning he and Jeremy Howard pioneered:
Abstract: Inductive transfer learning has greatly impacted computer vision, but existing approaches in NLP still…arxiv.org
What Your Writing Says About You
Visualize some of the common NLP tasks: part of speech tagging, sentiment analysis, complexity of writing.
An Overview of the FastAI Library
This is a mind map that help you navigate the FastAI library. Should be quite helpful when you want to customize things or extract some parts to your own framework.