Overfitting happens when machine learning identifies a pattern in the data that does not actually exist.
A simple example is finding a pattern in the Bible or Moby Dick that appears to predict future events. The learner (whether human or machine) starts to see patterns in the sea of data even though they don’t really exist.
The book goes in depth about prevention and detection mechanisms for overfitting.
The author describes the five most common camps in machine learning:
To the author, machine learning will eventually arrive at a place where we take the best approaches and implementations from all five camps and unite them into a master algorithm.
Machine learning still includes a lot of guess and check, which is probably why we see five distinct machine learning camps.
The author describes a future where everything is pre-optimized for us. Your matching algorithm(s) will result in you meeting only the people you’re likely to befriend, foods you’re likely to enjoy, and experiences you’re likely to repeat.
Bizarrely, the author presents this as a positive. A world perfectly optimized for each of us.
To my mind, this sounds like both boring dystopia and insanely inhumane.
How strongly do I recommend The Master Algorithm?
7 / 10
For software engineers, The Master Algorithm is a solid introduction to the landscape of machine learning. After reading this book, I feel like I know enough to ask useful and hopefully meaningful questions.