# Looped GNNs Beyond Fixed Depth: Graph Size Generalization and Parameter-Efficient Adaptation

DevFeed: [Looped GNNs Beyond Fixed Depth: Graph Size Generalization and Parameter-Efficient Adaptation](<https://devfeed.tech/articles/looped-gnns-beyond-fixed-depth-graph-size-generalization-and-parameter-efficient-adaptation-93241.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.24112>)

Author: Zhiyang Wang, Zhishang Luo, Yusu Wang

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [peft](<https://devfeed.tech/topics/peft.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [application](<https://devfeed.tech/tags/application.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [effects](<https://devfeed.tech/tags/effects.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gnn](<https://devfeed.tech/tags/gnn.md>), [graph](<https://devfeed.tech/tags/graph.md>), [information](<https://devfeed.tech/tags/information.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [processor](<https://devfeed.tech/tags/processor.md>), [radius](<https://devfeed.tech/tags/radius.md>), [reuse](<https://devfeed.tech/tags/reuse.md>), [sample](<https://devfeed.tech/tags/sample.md>), [size](<https://devfeed.tech/tags/size.md>), [small](<https://devfeed.tech/tags/small.md>), [sufficient](<https://devfeed.tech/tags/sufficient.md>)

## AI overview

Looped GNNs extend information reach to larger graphs by repeatedly applying a GNN processor pretrained on smaller graphs, without retraining. The paper studies tasks with a shared local iterative computation whose effective dependency radius grows with graph size, and categorizes six schemes to distinguish depth extension, fine-tuning, and freeze-and-steer adaptation. Its analysis gives fixed-depth information limits and conditions for accurate iterative reuse. For adaptation, Loop-Freeze-Steer (LFS) freezes the pretrained processor and trains a small controller to guide repeated application, using fewer trainable parameters than full fine-tuning. Finite-sample analysis provides conditions under which adaptation can improve larger-graph performance. Experiments on synthetic tasks and chip placement evaluate these approaches and their accuracy-trainable-parameter trade-off.

## Source excerpt

arXiv:2610.24112v1 Announce Type: new Abstract: Generalizing from small graphs to larger ones can require computation beyond the depth of a neural model used during training. We study target tasks that admit a common local iterative computation and whose effective dependency radius grows with graph size. The central idea is to learn a local update rule through GNN pretraining, then repeatedly apply the same learned processor beyond the training depth to extend information reach on larger graphs without retraining. Calling such parameter-shared architecture a \emph{looped GNN}, we categorize six schemes for looped GNNs to separate the effects of depth extension, fine-tuning, and freeze-and-steer adaptation. Our analysis establishes fixed-depth information limits and sufficient conditions under which accurate iterative reuse overcomes them. To improve beyond unadapted reuse, we introduce \emph{Loop-Freeze-Steer} (LFS), which freezes the pretrained processor and learns a small controller to steer its repeated application, using much fewer trainable parameters than full fine-tuning. Our finite-sample analysis gives sufficient conditions for adaptation to improve performance on larger graphs. Experiments on synthetic tasks and chip placement evaluate depth extension and adaptation, highlighting LFS's accuracy--trainable-parameter trade-off for size generalization.