> For the complete documentation index, see [llms.txt](https://zedive.gitbook.io/project-l/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://zedive.gitbook.io/project-l/part-3/advanced_topics/machine_learning/learning-theory/consistency_model.md).

# Consistency Model

## Definition

Let's start with this rather unrealistic but intuitive model. In this model, the prediction rule (from a set of examples) should be consistent with their observed labels. To formally state this, we say

a concept class $$c \in C$$ is learnable in the consistency model if $$\exists$$ an algorithm A such that given any set of labeled example $$(x\_1, y\_1), ..., (x\_m, y\_m)$$ where $$x\_i \in X$$ and $$y\_i \in {0,1}$$, finds a concept $$c \in C$$ so that $$c(x\_i) = y\_i;\forall i$$ (consistent with the examples), or says there is no such concept.

## Examples

Boolean Logic: monotone conjunctions, monotone disjunctions, conjunctions can be reduced to monotone conjuctions. They all can be proved learnable under the consistency model. Geometry: rectangles, half spaces More Boolean: K-CNF, DNF

## Problems with the consistency model

This model doesn't say anything about how the concept that the algorithm learns can generalize to new data. It seems unrelated to what we mean by "learning". We need a new model.