— In the News —
It's no secret that tech companies mine our personal data and come up with clever uses for it to generate lots of cash. For individuals, the result is targeted advertising and free, useful services, which most people don't seem to mind giving up some data for. But some nascent practices in the offline world are kind of creepy and dig deeper than we ever imagined.
"The next digital frontier is here, and it’s AI. "
— Tools and Techniques —
I'm a huge fan of mind mapping for idea generation and organization. If you like mind maps, this is a nice resource that covers the building blocks of machine learning. And if you're not familiar with mind maps, this is a great place to start to see how they're useful.
Why you should always carry a healthy dose of skepticism in your back pocket.
Awesome end-to-end tutorial for analyzing cryptocurrency markets. Starts with how to set up your environment and continues with retrieving pricing data, data wrangling, visualization, correlation analysis and lots of ideas for next steps.
Simple, plain-English explanations accompanied by math, code, and real-world examples.
— Deep Learning —
This collection of TensorFlow tutorials and best practices covers a variety of practical topics such as prototyping, multi-GPU processing, debugging, numerical stability and implementations of common operations. This is very well-written and worth bookmarking.
Great exploration of the history of automated music generation. From a technical perspective, this is pretty high level but the sound clips and descriptions create a great narrative. There are also lots of links throughout the article if you're interested in the algorithmic details.
— Resources —
The MNIST dataset contains a lot of handwritten digits and is commonly used for benchmarking machine learning algorithms. But researchers say that MNIST is too easy, it's overused, and it's outdated. Here's a drop-in replacement for the original MNIST.
— Data Viz —
Here's a super useful R package for making your production graphics more widely accessible. That's more important than you might think. Up to 8% of some populations are affected by various types of color blindness. If you're working with Python, check out colorspacious.
— In Case You Missed It —
Be sure to catch the most popular links from last week's issue...