This week in 'look what I built with scikit learn':
Almost certainly my first link to a piece in Confectionary News: reader and dear old friend Brianna points to a piece outlining Mars Co.'s investment in satellite technology designed to help small, independent cocoa farmers better predict weather patterns and allocate limited resources. It's a rather good example of the 'brand actions' we hear so much about and see so scarcely.
This author acknowledges that Mars Co. almost certainly does all manner of horrible things, and that you need not flood his inbox with examples.
* * *
Much of the language around the workplaces of the future has focused on the promise of symbiotic relationships between machines and humans — the idea of 'man plus machine' has become its own tired trope in some circles. Then, this week, in IEEE Spectrum:
Many robots lack a humanoid appearance and thus may not be able to make use of such cues, yet it is still critical that such appearance-constrained robots communicate their intentions to users. I think automobiles are a good example of this where we have a technology that we commonly use to communicate to other people, but in a different manner than traditional social communication.
* * *
Irving Wladawsky-Berger writes for the grand old web: Typepad, rich with links, still uses canonical categories. He's also a joy to read. This week's post is especially good, citing a paper by professors Erik Brynjolfsson and Tom Mitchell, identifying eight criteria for tasks that are especially well-suited to machine learning applications:
- Learning a function that maps well-defined inputs to well-defined outputs
- Large (digital) data sets exist or can be created containing input-output pairs
- The task provides clear feedback with clearly definable goals and metrics
- No long chains of logic or reasoning that depend on diverse background knowledge or common sense.
- No need for detailed explanation of how the decision was made
- A tolerance for error and no need for provably correct or optimal solutions
- The phenomenon or function being learned should not change rapidly over time
- No specialized dexterity, physical skills, or mobility required
Strong recommendation that you spend 6-7 minutes with the entire piece.
* * *
Have a look at Abhishek Singh's work, applying TensorFlow to training an Alexa to respond to sign language. Too often we allow the superintelligence narrative to overshadow the mundane brilliance of small, perfect ideas like this.