Research · Partially shipped

UTML — A Theory of Machine Learning

Chris Royse

What this paper covers

UTML models learning itself as a small set of factors — ΔP, ΔK, ΔΩ, ΔΞ — describing a move from entropy toward coherence. It is the learning theory the platform's own pedagogy engine is built on: the predict-reveal-explain loop, knowledge tracing, and spaced review are instances of UTML's account of how a learner's understanding becomes more coherent over time.

What the paper argues

UTML — A Theory of Machine Learning takes the same posture as the Calculus of Association and turns it on learning itself. It models learning as the interplay of a small set of factors — ΔP, ΔK, ΔΩ, ΔΞ — describing a directional move from entropy toward coherence: a learner (human or machine) whose understanding starts scattered and becomes organized.

This is not just theory the site cites — it is the theory the site runs on. The pedagogy engine’s predict → reveal → explain-back → mastery-check loop, its knowledge tracing, and its spaced review are all instances of UTML’s account of coherence increasing over time. The platform teaches the way it does because of this paper. For the factor-by-factor map from this theory to the buttons you actually click, see How this site teaches.

The bridge into Concepts

UTML closes the loop: the same measured, grounded posture that Calyx applies to meaning, the platform applies to teaching meaning.

Step down into the working vocabulary

Use these concept pages as the practical path into the paper's terms. Each one opens a focused explanation, demo, or source-backed learning surface.