1st Workshop on

Graphs Across AI: From Structural Reasoning to Foundation Models

Held as a part of the IEEE World Congress on Computational Intelligence (WCCI '26)

📆 26 June 2026, Maastricht 🇳🇱

Latest News

March 27, 2026 - The call for papers has now closed, with 14 articles submitted. We thank all the authors for their contributions and wish them the best of luck!
April 17, 2026 - Notifications have been sent to the authors. A total of 7 papers have been accepted, resulting in an acceptance rate of 50%.
May 27, 2026 - It has been officially announced that Davide Bacciu will be the invited speaker of the 1st GRAAI Workshop. We are delighted to welcome him and look forward to his talk!
Jun 4, 2026 - The workshop program has been published.

Abstract

Graphs provide a universal language for representing relationships, dependencies, and structural patterns across diverse domains. Their flexibility enables them to bridge data, models, and knowledge, establishing a connection between symbolic reasoning, statistical learning, and neural computation. While Graph Neural Networks (GNNs) have popularized graph-based learning, the role of graphs in Artificial Intelligence (AI) extends far beyond this paradigm. Graphs underpin computational structures in neural architectures, capture relational dependencies in attention mechanisms, organize knowledge through Knowledge Graphs (KGs), and model interactions among agents in social, multimodal, and dynamic environments.

This workshop aims to explore the expanding role of graph-based reasoning and representation across AI, moving beyond GNNs toward structure-aware learning both for and with foundation models. It seeks to foster discussion among researchers investigating how graphs can unify reasoning, learning, and computation across scales and modalities.

The goal is to consolidate theoretical, methodological, and application-oriented perspectives, strengthening the conceptual and practical connections between graphs and AI.

Topics of interest include, but are not limited to:

Graphs in neural architectures: modeling dependencies and computational structures in neural models.
Graph-based attention and representation learning: leveraging relational structures in attention mechanisms and embeddings.
Knowledge Graphs and reasoning: construction, integration, and use for symbolic and neuro-symbolic AI.
Graph-driven interpretability and explainability: using graphs to analyze and interpret model behavior.
Graphs for trustworthy retrieval: integration of large language models with knowledge graphs and graph-structured retrieval for reliable and explainable search.
Graphs in foundation model analysis: representing attention patterns and latent spaces as graphs to study token and patch relationships, pruning, and coarsening.
Graph-based Retrieval-Augmented Generation (RAG): enhancing retrieval, grounding, and response synthesis through graph-structured memory and knowledge integration.
Temporal, dynamic, and streaming graphs: modeling evolving structures for long-term memory and continual learning.
Graphs in multi-agent and multi-modal systems: representing interactions and coordination across agents, modalities, and environments.
Neuro-symbolic integration: connecting graph-based symbolic reasoning with neural computation.
Large-scale representation learning: scalable graph-based methods for learning over heterogeneous data.
Theoretical foundations: formal and mathematical frameworks linking graph theory with emerging AI paradigms.

Workshop Program

📆 Friday, June 26 - 📍 Mekong Room (2.18 MECC)

13:30 - 13:45 - Welcome and opening
13:45 - 14:45 - Keynote by Davide Bacciu - Designing Graph Neural Networks as dynamical systems
14:45 - 15:00 - S.M. Drammis, K. Srinivasan, N.A. Lynch, R. Ajemian - Building the dual graph of the activation regions in a deep neural network: what it means for generalization
15:00 - 15:15 - C. Buratti, M. Marchetti, F. Parlapiano, G. Quaglieri, D. Traini, D. Ursino, L. Virgili - Efficient Path-Coherent Retrieval via Branch Pruning in KG-RAG
15:15 - 15:30 - E. Izgorodin - Semantic level of detail for Knowledge Graphs: Discovering Abstraction Boundaries via Spectral Heat Diffusion
15:30 - 16:00 - Coffee Break
16:00 - 16:15 - S. Momtaz - Graph-structured Operator Learning with Conformal Calibration for Airport Network Forecasting
16:15 - 16:30 - A.D. Zenoozi, C. Aronis, A. Liotta, L. Cavallaro - Investigating Sparse Artificial Neural Network Topology through Similarity and Balance scores
16:30 - 16:45 - D. Sanchez, E. Dogdu, R. Choupani - Augmenting the unified Cybersecurity Knowledge Graph using RAG for Enhanced Threat Detection
16:45 - 17:00 - F. Bernabucci, G. Bonifazi, A. Rossi, L. Spalazzi - Graph Neural Networks for Multi-Entity Access Modeling in Zero Trust Architectures
17:00 - 17:15 - Closing and remarks

Invited Speaker

Davide Bacciu

Full Professor of Machine Learning at the Department of Computer Science, University of Pisa, where he leads the Pervasive AI Laboratory. His research focuses on the design of neural learning systems for graph-structured, temporal, and distributed data, with interests spanning graph representation learning, generative models, continual learning, and pervasive AI. He has co-authored over 250 research works on the field. He has coordinated major European research projects, including EIC Pathfinder EMERGE, and serves as Director of AI Research at Aptus.AI. He sits in the AdCom board of the IEEE Computational Intelligence Society, and previously served as Chair of the IEEE CIS Neural Network technical committee, as Senior Editor of the IEEE Transactions on Neural Networks and as Vice-President of the Italian Association for AI.

Designing Graph Neural Networks as dynamical systems

Most graph neural networks operate by exchanging information along the edges of a graph. This local mechanism of diffusion is a key reason for their effectiveness, but it also brings a fundamental architectural question: how can information propagate over sufficient distances to capture long-range dependencies without becoming diluted, oversmoothed, or lost?

In this talk, we reconsider the design of graph neural networks from a dynamical systems perspective. Instead of viewing message passing simply as a stack of generic layers, we interpret it as a neural flow in which propagation, conservation, and dissipation can be explicitly shaped. This viewpoint gives rise to principled architectures for both static and temporal graphs, while also providing theoretical tools to characterize when information is maintained and when it decays.

The keynote will highlight ideas with broad relevance for learning systems: how dynamics can guide model design, how stability and conservation can serve as architectural principles, and how theoretical guarantees can support the development of graph models that transmit information more reliably across complex data.

Organizers

Gianluca Bonifazi

Marche Polytechnic University, Italy

Claudio Gallicchio

University of Pisa, Italy

Barbara Hammer

Bielefeld University, Germany

Michele Marchetti

Marche Polytechnic University, Italy

Luca Virgili

Marche Polytechnic University, Italy

Program Committee

Alessia Amelio, Università G. D'annunzio
Inaam Ashraf, University of Bielefeld
Christopher Buratti, Marche Polytechnic University
Francesco Cauteruccio, University of Salerno
Gaetano Cimino, University of British Columbia
Stefano Cirillo, University of Salerno
Alessio Gravina, University of Pisa
Luca Hermes, University of Bielefeld
Benjamin Paassen, University of Bielefeld

Program Committee

Federica Parlapiano, Marche Polytechnic University
Vincenzo Pasquadibisceglie, University of Bari
Eliana Pastor, Polytechnic University of Turin
Giandomenico Solimando, University of Salerno
Serena Tardelli, IIT CNR
Matteo Tolloso, University of Pisa
Davide Traini, University of Modena and Reggio Emilia
Domenico Ursino, Marche Polytechnic University
Daniele Zambon, The Swiss AI Lab IDSIA & USI

Important Dates

Submission Deadline

Acceptance Notification

Camera-ready Papers

All submission deadlines are end-of-day in Anywhere on Earth (AoE) time zone.

Submission Instructions

Graphs Across AI invites research, industry, application contributions, and software demonstration submissions. Regular papers presenting original work are solicited. Moreover, discussion papers containing descriptions of results already published are also welcomed. There are two submission formats:

Regular papers (up to 12 pages + references). Original research works. Regular papers must be original and not simultaneously submitted to another journal or conference.
Discussion papers (up to 8 pages + references). Results and ideas of interest to the workshop audience, including extended abstracts of recent authors’ publications, papers currently under submission, position papers, system and application descriptions, and presentations of preliminary results.

Authors can submit and update submissions through the CMT system at the following link: https://cmt3.research.microsoft.com/GRAAI2026.

Paper selection will be based on originality, clarity, and technical quality. Submissions of papers must be in English, in PDF format, and formatted following the new CEUR-ART 1-column style, which is the style requested for the camera-ready preparation (Overleaf template).

Accepted regular and discussion papers will be included in the proceedings. Proceedings will be published on CEUR-WS.org. After acceptance, authors are required to re-submit the final PDF.

The Microsoft CMT service was used for managing the peer-reviewing process for this conference. This service was provided for free by Microsoft and they bore all expenses, including costs for Azure cloud services as well as for software development and support.

Google Sites

Report abuse