logic-gnn-2026
We are delighted to organise a meeting on logics and graph neural networks (GNNs) in Queen Mary University of London.
We observe a growing interest in combining logical and neural approaches, with GNNs proving to be a particularly well-suited AI architecture for such an integration. As a result, tight connections between logics and GNNs are attracting increasing attention, with current directions including:
- Expressive power of GNN architectures
- Verification of GNN models using formal methods
- Logical explainability and interpretability of GNNs
- Rule learning and knowledge discovery with GNN models
- Computational complexity of GNN architectures
The aim of this meeting is to bring together researchers from this growing, but still relatively young, area to discuss recent results, ongoing work, and future directions.
Venue
Queen Mary University of London, Mile End campus, London E1 4NS,
Maths Lecture Theatre—building number 4 on the campus map
Registration
To attend the event please fill the form - the registration will close on Tuesday 23 June.
Please notice that filling the form is sufficient to register, and no confirmations will be sent to attendants.
(Speakers and other invited researchers who have already filled out the separate form do not need to complete the above form.)
PROGRAM (subject to change)
Thursday 25 June 2026
9:45 – 10:00 Opening Talk
10:00 – 11:00
- Floris Geerts, A Logical View of GNN-Style Computation and the Role of Activation Functions, paper
- Veeti Ahvonen, On the Connections Between Graph Transformers and GNNs, paper
11:00 – 11:30 Coffee break
11:30 – 12:30
- David Tena Cucala, Practical Rule Extraction from Monotonic GNNs
- Przemysław Wałęga, Preservation Theorems for GNNs, paper 1, paper 2
12:30 – 14:00 Lunch break
14:00 – 15:30
- Stan Hauke, Aggregate-Combine-Readout GNNs Can Express Logical Classifiers Beyond the Logic C2, paper
- Maurice Funk, Daumantas Kojelis, Towards Understanding the Expressive Power of GNNs with Global Readout, paper
- Nicolas Troquard, About Logic and Aggregate-Combine(-Readout) Graph Neural Networks over Integers, paper 1, paper 2
15:30 – 16:00 Coffee break
16:00 – 16:30 Brian Shi, Graph Reasoning Agents at Neo4j: Integrating Graph Queries, Algorithms, and ML
16:30 – 17:30 Discussion panel: Floris Geerts, Antti Kuusisto, Carsten Lutz, François Schwarzentruber
Friday 26 June 2026
10:00 – 11:00
- Antti Kuusisto, Characterizing recurrent GNNs and NNs, paper 1, paper 2
- Eva Feng, The Correspondence between Bounded GNNs and Fragments of FOL, paper
11:00 – 11:30 Coffee break
11:30 – 12:30
- Jonni Virtema, Unifying approach to uniform expressivity of graph neural networks, paper
- Marco Sälzer, The Polynomial Counting Capabilities of Message Passing Neural Networks, paper
12:30 – 14:00 Lunch break
14:00 – 15:30
- Chia-Hsuan Lu, Decidability of Graph Neural Networks via Logical Characterizations, paper
- François Schwarzentruber, Verifying Quantized Graph Neural Networks is PSPACE-complete, paper
- Damian Heiman, Neural networks as fuzzy logic formulas, paper
15:30 – 16:00 Coffee break
16:00 – 17:30
- Hubie Chen, Optimally Rewriting Formulas and Database Queries: A Confluence of Term Rewriting, Structural Decomposition, and Complexity, paper
- Marc Roth, The Weisfeiler-Leman Dimension of Conjunctive Queries, paper
- Marek Černý, Restricted-Conjunction Logic Turns Message Passing into Multiplicity Automata, paper
Organisation
- Przemysław Wałęga, School of Electronic Engineering and Computer Science, Queen Mary University of London