Data Science articles I have read this week (w/c 1/11/21
This is a philosophical article not only about AI creating Beethoven's 10th symphony from drafts that were found but also about AI in general. I personally don’t like how the word AI is used universally mostly to describe machine learning algorithms. The author points out all the things that separate us from AI, especially feelings, emotions but also dreams and creativity as well as labour and trying hard. To an uneducated ear like mine the AI 10th symphony sounds like a classical piece of music. It’s not spectacular or particularly memorable but seems to flow coherently (there is a link to the video in the article).?
This article summarises machine learning specialities of selected big tech firms such as Facebook and Google. They have each specialised in slightly different domains of machine learning. OpenAI are hopeful to develop a new transformer model that might pass the Turning test. Meanwhile, Apple are focussing on federated learning that allows model training with decentralised data. There are plenty of links in the article to dig into the different domains.
领英推荐
I’m a big fan of SHAP, the charts are beautiful and explain complicated concepts like feature importance. However, just yesterday I tried to use it with a large data set and it would have taken too long to run which is such a shame. The article explains what SHAP is and how to use it to interpret machine learning model results. It’s most compatible with tree models but also works on other algorithms. These kind of interpretations of model results are especially valuable to ‘black box’ models and help make interpretability of machine learning models a reality.
The article discusses that current state of Data from the viewpoint of tooling, workflow and team structures. The author compares data teams and workflows to software engineering which is believed to have reached a state of efficient workflow and well organised teams to optimise this. In data, it is argued, team structure has not caught up with technological advancement of data tools yet. I especially like the reference to business expectations about timeliness of data work. You would not expect a software engineer to produce a product within a day and you probably wouldn’t want to be cause you know there will be lots of bug and it won’t have the functionality you require. Why don’t we think like this about data requests??