Tag
The paper demonstrates that recursive language models improve out-of-domain generalization by isolating subtasks, avoiding shortcuts used by standard Chain of Thought, and highlighting limitations in classical learning theory for true reasoning.
This paper critiques conventional depth truncation methods for evaluating recursive language models and introduces the Depth Control Protocol (DCP) to disentangle and isolate factors affecting depth utilization, improving evaluation accuracy.
This ICML paper introduces recursive models that recursively invoke themselves to solve subtasks in isolated contexts, proving they can surpass context-bounded autoregressive models for long-horizon reasoning. Experiments on SAT solving and Go game-tree search show improved accuracy with small active contexts.
Proposes interaction locality, a task-geometry-aware framework for measuring whether information flow in spatial reasoning models stays within local cells or crosses into global structure, and applies it to HRM, TRM, and MTU3D models on grid benchmarks and embodied 3D grounding.