E4 — Pattern Essays

Signal in the Noise#

Essay
1 min read
◆ The PatternFinding patterns is easy. The hard part is resisting them where there are none

Scatter a hundred random points on a plane. Stare long enough and you will see clusters, streaks, shapes. The human brain is a pattern-completion machine, built for a world where mistaking a shadow for a predator was safer than ignoring a predator. That wiring does not switch off when you look at data.

Every dataset is a blend of true signal and meaningless noise. A model trained too eagerly memorises the noise and calls it knowledge. That is the textbook definition of overfitting. The antidote is restraint: hold data back, cross-validate, penalise complexity, and accept that “I don’t know” is sometimes the most accurate answer.

The pattern here is a meta-pattern: the urge to see patterns can itself be the error. The discipline of statistics is, at its core, a set of tools for telling the difference.

Noise level 50%
A wave hiding in noise. Drag to reveal or bury it
What to rememberBefore trusting a pattern, ask whether you would still see it in fresh data you have not looked at yet.
References
[1] Silver, N. (2012). The Signal and the Noise: Why So Many Predictions Fail — but Some Don’t. Penguin Press.
[2] Foster, K. R. & Kokko, H. (2009). The evolution of superstitious and superstition-like behaviour. Proceedings of the Royal Society B, 276(1654), 31–37. doi:10.1098/rspb.2008.0981
[3] Hastie, T., Tibshirani, R. & Friedman, J. (2009). The Elements of Statistical Learning, Ch. 7: Model Assessment and Selection. Springer. doi:10.1007/978-0-387-84858-7

Last reviewed October 2026 · How this site checks its content · Report a mistake

Pattern bridge: Cross-validation is the practical guard against this, and learning curves let you see overfitting happen in real time. Searching many rules until one works is the trap of The Garden of Forking Paths.
Linked from 5
← Previous
The Long Tail
Open in the full reader, with the topic sidebar