E8 — Pattern Essays

The Threshold#

Essay2 min read
1 min read
◆ The PatternMany systems stay quiet for a long time — then change all at once

Add grains of sand to a pile, one at a time. For a long while, nothing dramatic happens. Then, at some unpredictable moment, a single grain triggers an avalanche. The pile was always close to collapse. The last grain gets the credit it did not deserve.

This is threshold behaviour, and it appears everywhere: a rumour that suddenly goes viral, ice that holds firm and then fractures, a neuron that fires only when input crosses a minimum. In each case, input and output are not proportional. Small changes accumulate invisibly until the threshold is crossed, and then the system snaps.

The sigmoid function is the mathematician’s version: nearly flat on both sides, steep in the middle. It describes the dose-response curve of a drug, the probability of a binary outcome in logistic regression, and the activation of a neuron. The threshold is not special — it is simply the midpoint of a curve that was always going to be steep somewhere.

Input level 30
The sigmoid — drag through the threshold
What to rememberSlow, invisible accumulation can end in a sudden snap — watch the approach to the line, not just the line itself.
References
[1] Bak, P., Tang, C. & Wiesenfeld, K. (1987). Self-organized criticality. Physical Review Letters, 59(4), 381–384. doi:10.1103/PhysRevLett.59.381
[2] Gladwell, M. (2000). The Tipping Point: How Little Things Can Make a Big Difference. Little, Brown and Company.
[3] Strogatz, S. H. (1994). Nonlinear Dynamics and Chaos. Addison-Wesley. Ch. 3: Bifurcations.
Pattern bridge: Logistic regression in the ML Lab is built on this very curve, and hypothesis testing uses a threshold (the p-value) to decide when evidence becomes belief.
← Previous
The Random Walk
Open in the full reader, with the topic sidebar →