|September 15 · Issue #57 · View online |
Hi and welcome to a new week in deep learning,
Happy hacking and reading. As always if you enjoy receiving this newsletter you can help us by sharing it with friends and colleagues.
See you next week!
| In search of the missing signals |
In this well-written overview, Giorgio Patrini explores the current state of unsupervised learning and shares his thoughts on some recent approaches and techniques. The way he explains the papers allows you to quickly jump into the field and get a feel for the different trends that came up recently. Definitely worth a look!
| We are the Google Brain team. We’d love to answer your questions (again) |
The Google Brain team has decided to do another AMA on Reddit, after last years one was very successful. They share their opinions on Pytorch, try to anticipate upcoming challenges and hurdles, e.g. unsupervised learning and solving multiple task with a single model, as well as insights into the daily life at Google Brain.
| AI can liberate humans to lead happier lives, if we get it right |
Another take on the heated discussions around the impact of AI on general life and the workplace of the future. This time a little more balanced and not the usual apocalyptic approach. A four-day working week seems like a great argument for AI.
| Video Object Segmentation — The Basics – Eddie Smolyansky – Medium |
This is the first of two articles on video object segmentation. It covers the basic concepts, existing datasets and last years main approaches, as well as a new dataset made by Visualead and focused on e-commerce. A great read if you want to get into video object segmentation or quickly need to get up to speed. Don’t forget the second article
for more advanced insights.
| Detecting Malicious Requests with Keras & Tensorflow |
This article describes the results of a recent hackathon, where the team built a machine learning system that tries to classify API requests as regular and malicious. Interestingly they opted for a NLP based approach and the resulting model seems quite efficient, but the lessons learned are interesting as well.
| Introducing Pytorch for fast.ai |
Fast.ai will begin using Pytorch instead of Keras and TensorFlow for their upcoming courses and explains the motivations behind the change. The dynamic aspect of the framework seems to speed up adoption quite heavily, so why not take a look?
| GitHub - fyu/drn: Dilated Residual Networks |
Speaking of Pytorch, this repository contains a collection of residual neural networks. They’re able to achieve good results on different datasets and may be useful for your next project.
| Build your own Machine Learning Visualizations with the new TensorBoard API |
Google has released a new API for TensorBoard that allows creating new visualizations using a plugin based system. An initial set of new plugins
using the API is already available and may be used as a starting point for your own implementations.
| Deep Learning Techniques for Music Generation - A Survey |
Whenever a specific field grows exponentially, it can become really hard to grasp all available approaches and put them in perspective. Thanks to this recent paper one can now get an overview of all recent works on the generation of music using deep learning.
| NIPS 2017 - Accepted Papers |
This years accepted papers have been announced and you can now get a sneak peek at what’s to come. Definitely a lot of interesting titles in there.