Welcome to Simple Machine Learning
Why this journal exists, and what to expect from each issue.
Most writing about machine learning falls into one of two ditches. The first is the textbook ditch, where every idea arrives wrapped in three layers of notation and the intuition is left as an exercise for the reader. The second is the marketing ditch, where everything is a paradigm shift and nothing is ever wrong.
This journal aims for the road in the middle. Each issue is a single essay on something specific — a loss function, a failure mode, a small experiment — written plainly enough that you can read it on a phone, but technically exact enough that you can act on it.
What you can expect
- One essay per issue. No filler.
- Code that runs. If a snippet appears, it has been executed.
- Honest failure modes. Where a method breaks, the essay says so.
The model is wrong. The model is always wrong. The art is knowing by how much.
What you will not find
No newsletter pop-ups. No cookie banner theatrics. No "subscribe to keep reading." The whole archive is free to read, and the source is a folder of Markdown files.
def loss(y, yhat):
return ((y - yhat) ** 2).mean()
Thanks for stopping by. The next issue is already in the press.