Abstract:In the semi-supervised setting where labeled data are largely limited, it remains to be a big challenge for message passing based graph neural networks (GNNs) to learn feature representations for the nodes with the same class label that is distributed discontinuously over the graph. To resolve the discontinuous information transmission problem, we propose a control principle to supervise representation learning by leveraging the prototypes (i.e., class centers) of labeled data. Treating graph learning as a discrete dynamic process and the prototypes of labeled data as "desired" class representations, we borrow the pinning control idea from automatic control theory to design learning feedback controllers for the feature learning process, attempting to minimize the differences between message passing derived features and the class prototypes in every round so as to generate class-relevant features. Specifically, we equip every node with an optimal controller in each round through learning the matching relationships between nodes and the class prototypes, enabling nodes to rectify the aggregated information from incompatible neighbors in a graph with strong heterophily. Our experiments demonstrate that the proposed PCGCN model achieves better performances than deep GNNs and other competitive heterophily-oriented methods, especially when the graph has very few labels and strong heterophily.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problem of how to effectively learn node feature representations in graph neural networks (GNNs) in a semi - supervised setting when the labeled data is very limited. Specifically, the article addresses the following challenges: 1. **Discontinuous information transfer problem**: For nodes that are distributed in different regions of the graph but belong to the same category, traditional message - passing - based GNNs have difficulty effectively spreading information. As the topological distance increases, the influence of the labeled nodes gradually weakens. 2. **Over - smoothing problem**: Increasing the number of message - passing layers will lead to over - smoothing of node feature representations, that is, the representations of nodes become too similar, mainly determined by the graph structure rather than their class labels. 3. **Negative smoothing effect in heterogeneous graphs**: In heterogeneous graphs, adjacent nodes often come from different categories. The traditional message - passing mechanism will force the feature representations of these nodes to converge to the average value of the local neighborhood, thus deteriorating the effect of representation learning. To solve these problems, the authors propose a new control principle - **Pinning Control**, which supervises representation learning by using the prototypes of the labeled data (i.e., the class centers). Specifically, the authors regard graph learning as a discrete dynamic process, and use the prototypes of the labeled data as the "desired" class representations. They design a learning feedback controller by borrowing the idea of pinning control in automatic control theory to minimize the difference between the features generated by message passing and the class prototypes, thereby generating class - related features. By introducing pinning control, this method can adjust the aggregated information of nodes in each message - passing process and correct the noisy information from incompatible neighbors, especially in strongly heterogeneous graphs. Experimental results show that the proposed PCGCN model outperforms deep GNNs and other competing heterogeneity - oriented methods in the case of very limited labels and strongly heterogeneous graphs. ### Main contributions 1. **Introduction of pinning control**: A new graph representation learning framework is proposed. By introducing the pinning control method, a learning feedback controller is used to supervise representation learning and directly transfer class - related information to each node. 2. **End - to - end model**: An end - to - end model is developed to learn the representations of class prototypes and dynamically select and update the pinning controllers during the message - passing process, so that unlabeled nodes can be directly supervised by the prototypes of potential classes, solving the problem of long - distance message passing. 3. **Extensive experimental verification**: Extensive experiments have been carried out on a variety of real - world graph datasets, proving that this method significantly improves the performance of traditional message - passing GCNs and performs excellently among the state - of - the - art GNN models with different message - passing schemes, especially when the network labels are limited. ### Summary This paper effectively solves the problems of discontinuous information transfer, over - smoothing and negative smoothing effects faced by graph neural networks when learning node feature representations in a semi - supervised setting by introducing the pinning control mechanism, especially performing well in heterogeneous graphs.

Steering Graph Neural Networks with Pinning Control

On the Pinning Strategy of Complex Networks

Pinning Control on Complex Networks Via Identifying Communities

p-Laplacian Based Graph Neural Networks

Global Pinning Synchronization with PI Controller in General Complex Directed Networks

Discovering the Representation Bottleneck of Graph Neural Networks

Simplified PCNet with Robustness

Adaptive Group Consensus Of Multi-Agent Networks Via Pinning Control

Learning Decentralized Flocking Controllers with Spatio-Temporal Graph Neural Network

Graph Neural Networks with Heterophily

Graph Pointer Neural Networks

Graph Stochastic Neural Networks for Semi-supervised Learning.

Deconfounded Training for Graph Neural Networks

Graph Neural Networks with Feature and Structure Aware Random Walk

Learning on Graphs under Label Noise

Synchronization Control for Coupled Delayed Neural Networks With Time-Varying Coupling Via Markov Pinning Strategy

Control the GNN: Utilizing Neural Controller with Lyapunov Stability for Test-Time Feature Reconstruction

Semi-Supervised Mixture Learning for Graph Neural Networks With Neighbor Dependence

Beyond Homophily in Graph Neural Networks: Current Limitations and Effective Designs

DGNN: Decoupled Graph Neural Networks with Structural Consistency between Attribute and Graph Embedding Representations