Nested Shared-Operator Recursion on ARC-AGI-1: An Empirical Study of HRM, TRM and URM

Key points

• We investigate whether adding a persistent intermediate state that integrates repeated updates of a faster state through the same operator improves predictive performance relative to each model family's baseline at matched parameter counts.1
• We define this organisation operationally and implement it within three architectures: the Hierarchical Reasoning Model (HRM), Tiny Recursive Model (TRM) and Universal Reasoning Model (URM).1
• Using the official codebases on ARC-AGI-1, we compare each baseline with its parameter-matched nested variant.1
• We evaluate held-out queries from tasks whose demonstrations were available during training.1

Sources (1)

• [1]Nested Shared-Operator Recursion on ARC-AGI-1: An Empirical Study of HRM, TRM and URM
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 04:20 PM

  Recursive reasoning models refine persistent internal states by repeatedly applying learned operators.

  We investigate whether adding a persistent intermediate state that integrates repeated updates of a faster state through the same operator improves predictive performance relative to each model family's baseline at matched parameter counts.

Extractive summary: sentences quoted from the sources.