News

Przemyslaw Wałęga has had two full papers (and 2 further extended abstract papers) accepted for the 23rd International Conference on Principles of Knowledge Representation and Reasoning, July 20-23, 2026 - Lisbon, Portugal. KR is the main conference paper on knowledge representation and reasoning, which takes place as part of FLoC.

The first paper is about the expressive power of graph neural networks, depending on the form of the aggregation function (used to aggregate information about neighbours in a graph), being SUM, MEAN, MAX, or arbitrary. The landscape turns out to be non-trivial and depending on whether GNN has "global readout" (ACR-GNN) and whether it is simple (uses a single perceptron as a combination function) or not we obtain different results.

The second KR paper is a result of collaboration with Mathijs van Noort who visited QMUL in 2025. They have showed that instead of introducing new solvers for temporal logic programming, we can easily extend standard (non-temporal) logic programming solvers. Due to access to mature non-temporal solvers, this approach is competitive even with dedicated temporal reasoners. This is surprising and suggests practical usefulness of our approach.

References
S. P. Hauke, P. A. Wałęga, How Aggregation Functions Affect the Uniform Expressiveness of Graph Neural Networks, KR, 2026

M. van Noort, P. A. Wałęga, Efficient Temporal Reasoning with Non-Temporal Engines: Embedding DatalogMTL into Datalog, KR, 2026