— In the News —
The gold rush started last October when Christie's sold an algorithm generated print for $432,500. More recently, an AI artist had its own show at a gallery in Chelsea. There's definitely a lot of interest here but is AI art really all that interesting? This longread in the Atlantic explores this burgeoning industry with links to artwork so you can judge for yourself.
— Insight —
A team of specialists works well in environments where the organization knows exactly what needs to be done and execution can be managed like an assembly line. This article by Eric Colson explores why that's rarely the case in data science and how specialization can get in the way.
— Tools and Techniques —
Nice post that starts by showing how every matrix is a graph. From there, it's a visual tour of matrix operations and probabilities. Great read!
In this excerpt from his model explainability course, Dan Becker outlines the types of things that the very best data scientists are able to discern about their models and why that information is useful. This post also sparked a worthwhile discussion on Hacker News.
Great interactive article on the Distil site that introduces a new technique for visualizing how decision-making happens in a neural network. It's a long read but it's compelling all the way through.
By default, Jupyter Notebooks are unnamed, have no markdown cells, and no imports. Since people are notoriously bad at changing default settings, why not encourage better practices? This simple extension gently nudges you to create better notebooks.
Building NLP systems in a complex domain like health care is hard. Not only do these systems require broad domain knowledge, every sub-specialty and form of communication is fundamentally different. In this post, David Talby outlines common issues and the lessons he's learned over 7 years of building NLP systems in health care.
Vettery specializes in tech roles and is completely free for job seekers. Interested? Submit your profile, and if accepted onto the platform, you can receive interview requests directly from top companies growing their data science teams.
— Resources —
This curated list of machine learning interpretability resources is definitely worthy of its "awesome" moniker. Includes a blueprint of use-cases, software examples, tutorials, packages, books, papers, etc.
— Data Viz —
"Logo design" may not sound interesting but this post describes the logo for the newly formed Data Visualization Society. The logo changes dynamically according to member skills and it's unlike any logo you've ever seen.