Roundup 11/11/2016

Machine learning (ML) and deep learning (DL) content from the past 24 hours.
TensorFlow @ 1
A year ago, Google released TensorFlow to open source. Since then, the company has added distributed computing features, support for iOS and a managed service in the cloud. TensorFlow is now the most popular machine learning project on GitHub, and it has the largest active contributor community among deep learning projects listed on OpenHub. Google celebrates.
Fundamentals
— Taylor Hall asks: why machine learning models often fail to learn.
— Shuvro Sarkar explains how to use machine learning in the enterprise.
Methods and Techniques
— Sebastian Ruder surveys gradient descent optimization algorithms. If that sounds esoteric, read the article; optimization is the DNA of machine learning.
— In Engineering and Technology, Chris Edwards describes the role of neural networks in a new generation of robots.
— Matthew Mayo explains GPU-based parallelism and the CUDA framework.
Software
— Kareem Anderson describes some of the machine learning services available in Microsoft Azure. He omits Microsoft R Server for some reason.
Applications
— Researchers use TensorFlow to detect Sea Cows in ocean images. It’s part of a project to protect the endangered animals.
— A Fujitsu application built with deep learning can recognize handwritten Chinese characters with 96.3% accuracy — better than humans.
— Janko Roettgers describes how media companies use AI to spam you.
Companies
— RiskIQ raises $30.5 million in new funding. The company markets software that uses machine learning to help organizations defend against cyber threats.
Weekend Reads
Here are three good books on machine learning, all with great ratings on Amazon.
Kuhn and Johnson: Applied Predictive Modeling.
Flach: Machine Learning.
Lantz: Machine Learning with R.