curio

philosophy

A perfect match for old data can be a bad guide to new data

Imagine dots showing measurements on a chart. You draw a twisting line through every dot, so it fits the old results perfectly.

But some twists may copy random quirks rather than a pattern that repeats. A new measurement can land far from where the line predicts.

This is overfitting: copying old examples too closely. A simpler line can sometimes predict new results better, even if it misses some old dots.

Source: arxiv.org

Get Curio — more like this

Short reads that go somewhere. No feed to scroll forever.