Rémi Bourgerie

Remi Bourgerie, Sarunas Girdzijauskas, Viktoria Fodor

Deep Graph Neural Networks (GNNs) are essential for capturing complex dependencies in graph-structured data. However, scaling GNNs to depth remains challenging, as stacking layers leads to representation collapse and diminishing sensitivity due to repeated aggregation. While Neural Sheaf Diffusion (NSD) provides strong theoretical guarantees against such collapse, these guarantees do not translate to practice: as depth increases, the disagreement signal of the sheaf Laplacian vanishes, limiting the contribution of deeper layers. We identify mechanisms that hinder NSD effectiveness at depth and propose \emph{Deep Neural Sheaf Diffusion} (DNSD), which replaces the sheaf Laplacian with a sheaf adjacency operator to maintain informative signals across layers. This is complemented by normalization, odd nonlinearities, and gating. To provide a principled explanation of the expected performance improvement, we contrast sheaf diffusion to graph attention mechanisms, highlighting that DNSD replaces scalar attention scores with matrix-valued edge functions and normalizes node representations rather than attention scores. We demonstrate empirically that DNSD effectively utilizes deep aggregation in graph tasks, outperforming GNN and NSD baselines with up to 30pp accuracy on synthetic long-range datasets, and consistently outperforming them on real-world benchmarks. These results position sheaf-based architectures as a promising building block for graph foundation models by supporting effective deep architectures.

R. Bourgerie, T. Zanouda

5G and Beyond Networks become increasingly complex and heterogeneous, with diversified and high requirements from a wide variety of emerging applications. The complexity and diversity of Telecom networks place an increasing strain on maintenance and operation efforts. Moreover, the strict security and privacy requirements present a challenge for mobile operators to leverage network data. To detect network faults, and mitigate future failures, prior work focused on leveraging traditional ML/DL methods to locate anomalies in networks. The current approaches, although powerful, do not consider the intertwined nature of embedded and software-intensive Radio Access Network systems. In this paper, we propose a Bi-level Federated Graph Neural Network anomaly detection and diagnosis model that is able to detect anomalies in Telecom networks in a privacy-preserving manner, while minimizing communication costs. Our method revolves around conceptualizing Telecom data as a bi-level temporal Graph Neural Networks. The first graph captures the interactions between different RAN nodes that are exposed to different deployment scenarios in the network, while each individual Radio Access Network node is further elaborated into its software (SW) execution graph. Additionally, we use Federated Learning to address privacy and security limitations. Furthermore, we study the performance of anomaly detection model under three settings: (1) Centralized (2) Federated Learning and (3) Personalized Federated Learning using real-world data from an operational network. Our comprehensive experiments showed that Personalized Federated Temporal Graph Neural Networks method outperforms the most commonly used techniques for Anomaly Detection.

Ankit Grover, Lodovico Giaretta, Rémi Bourgerie et al.

The integration of Graph Neural Networks (GNNs) with Large Language Models (LLMs) has emerged as a promising paradigm for Graph Question Answering (GraphQA). However, effective methods for encoding complex structural information into the LLM's latent space remain an open challenge. Current state-of-the-art architectures, such as G-Retriever, typically rely on standard GNNs and aggressive mean pooling to compress entire graph substructures into a single token, creating a severe information bottleneck. This work mitigates this bottleneck by investigating two orthogonal strategies: (1) increasing the bandwidth of the graph-to-LLM interface via multi-token pooling, and (2) enhancing the semantic quality of the graph encoder via global attention mechanisms. We evaluate a suite of hierarchical pruning and clustering-based pooling operators including Top-k, SAGPool, DiffPool, MinCutPool, and Virtual Node Pooling (VNPool) to project graph data into multiple learnable tokens. Empirically, we demonstrate that while pooling introduces significant instability during soft prompt tuning, the application of Low-Rank Adaptation (LoRA) effectively stabilizes specific hierarchical projections (notably VNPool and pruning methods), though dense clustering operators remain challenging. This stabilization allows compressed representations to rival full-graph baselines (achieving ~73% Hit@1 on WebQSP). Conceptually, we demonstrate that a Graph Transformer with VNPool implementation functions structurally as a single-layer Perceiver IO encoder. Finally, we adapt the FandE (Features and Edges) Score to the generative GraphQA domain. Our analysis reveals that the GraphQA benchmark suffers from representational saturation, where target answers are often highly correlated with isolated node features. The implementation is available at https://github.com/Agrover112/G-Retriever/tree/all_good/