Procedural Generation of Diverse Examples for the Abstraction and Reasoning Corpus

This work presents a method to procedurally generate diverse examples for the Abstraction and Reasoning Corpus (ARC) tasks, enabling fundamental experiments on sample-efficient learning.