Lost in Serialization: Invariance and Generalization of LLM Graph Reasoners

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Herbst, Daniel, Karbevska, Lea, Kumar, Divyanshu, Ahuja, Akanksha, Nasrabadi, Fatemeh Gholamzadeh, Frasca, Fabrizio
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917106335875072
author Herbst, Daniel
Karbevska, Lea
Kumar, Divyanshu
Ahuja, Akanksha
Nasrabadi, Fatemeh Gholamzadeh
Frasca, Fabrizio
author_facet Herbst, Daniel
Karbevska, Lea
Kumar, Divyanshu
Ahuja, Akanksha
Nasrabadi, Fatemeh Gholamzadeh
Frasca, Fabrizio
contents While promising, graph reasoners based on Large Language Models (LLMs) lack built-in invariance to symmetries in graph representations. Operating on sequential graph serializations, LLMs can produce different outputs under node reindexing, edge reordering, or formatting changes, raising robustness concerns. We systematically analyze these effects, studying how fine-tuning impacts encoding sensitivity as well generalization on unseen tasks. We propose a principled decomposition of graph serializations into node labeling, edge encoding, and syntax, and evaluate LLM robustness to variations of each of these factors on a comprehensive benchmarking suite. We also contribute a novel set of spectral tasks to further assess generalization abilities of fine-tuned reasoners. Results show that larger (non-fine-tuned) models are more robust. Fine-tuning reduces sensitivity to node relabeling but may increase it to variations in structure and format, while it does not consistently improve performance on unseen tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10234
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Lost in Serialization: Invariance and Generalization of LLM Graph Reasoners
Herbst, Daniel
Karbevska, Lea
Kumar, Divyanshu
Ahuja, Akanksha
Nasrabadi, Fatemeh Gholamzadeh
Frasca, Fabrizio
Machine Learning
Artificial Intelligence
While promising, graph reasoners based on Large Language Models (LLMs) lack built-in invariance to symmetries in graph representations. Operating on sequential graph serializations, LLMs can produce different outputs under node reindexing, edge reordering, or formatting changes, raising robustness concerns. We systematically analyze these effects, studying how fine-tuning impacts encoding sensitivity as well generalization on unseen tasks. We propose a principled decomposition of graph serializations into node labeling, edge encoding, and syntax, and evaluate LLM robustness to variations of each of these factors on a comprehensive benchmarking suite. We also contribute a novel set of spectral tasks to further assess generalization abilities of fine-tuned reasoners. Results show that larger (non-fine-tuned) models are more robust. Fine-tuning reduces sensitivity to node relabeling but may increase it to variations in structure and format, while it does not consistently improve performance on unseen tasks.
title Lost in Serialization: Invariance and Generalization of LLM Graph Reasoners
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.10234