Statistical Guarantees for Reasoning Probes on Looped Boolean Circuits

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kratsios, Anastasis, Livieri, Giulia, Neuman, A. Martina
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910274796126208
author Kratsios, Anastasis
Livieri, Giulia
Neuman, A. Martina
author_facet Kratsios, Anastasis
Livieri, Giulia
Neuman, A. Martina
contents We study the statistical behavior of reasoning probes in a stylized model of iterative computation inspired by neural algorithmic reasoning. The underlying computation is given by a looped Boolean circuit whose graph is a perfect $ν$-ary tree ($ν\ge 2$), with outputs recursively fed back as inputs across computation rounds. A probe observes a sampled subset of internal nodes and seeks to infer the latent operation at each node, represented as a probability distribution over a finite set of admissible Boolean gates. This partial observability induces a transductive generalization problem on a structured computation graph. We show that when the probe is parameterized by a graph convolutional network and queries $N$ nodes, the worst-case generalization error decays at the optimal rate $\mathcal{O}(\sqrt{\log(2/δ)}/\sqrt{N})$ with probability at least $1-δ$. Our analysis combines metric embedding techniques with tools from optimal transport. A key insight is that this rate is achievable independently of the size of the computation graph, enabled by a low-distortion one-dimensional snowflake embedding of the induced graph metric. These results highlight a geometric mechanism underlying statistical efficiency in probing structured, iterative computations.
format Preprint
id arxiv_https___arxiv_org_abs_2602_03970
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Statistical Guarantees for Reasoning Probes on Looped Boolean Circuits
Kratsios, Anastasis
Livieri, Giulia
Neuman, A. Martina
Machine Learning
Neural and Evolutionary Computing
Metric Geometry
Statistics Theory
We study the statistical behavior of reasoning probes in a stylized model of iterative computation inspired by neural algorithmic reasoning. The underlying computation is given by a looped Boolean circuit whose graph is a perfect $ν$-ary tree ($ν\ge 2$), with outputs recursively fed back as inputs across computation rounds. A probe observes a sampled subset of internal nodes and seeks to infer the latent operation at each node, represented as a probability distribution over a finite set of admissible Boolean gates. This partial observability induces a transductive generalization problem on a structured computation graph. We show that when the probe is parameterized by a graph convolutional network and queries $N$ nodes, the worst-case generalization error decays at the optimal rate $\mathcal{O}(\sqrt{\log(2/δ)}/\sqrt{N})$ with probability at least $1-δ$. Our analysis combines metric embedding techniques with tools from optimal transport. A key insight is that this rate is achievable independently of the size of the computation graph, enabled by a low-distortion one-dimensional snowflake embedding of the induced graph metric. These results highlight a geometric mechanism underlying statistical efficiency in probing structured, iterative computations.
title Statistical Guarantees for Reasoning Probes on Looped Boolean Circuits
topic Machine Learning
Neural and Evolutionary Computing
Metric Geometry
Statistics Theory
url https://arxiv.org/abs/2602.03970