Published in Data Mining and Knowledge Discovery, 2026
OpenGraphXAI benchmarks, with a method to guarantee ground truth learnability and uniqueness.
Recommended citation: M. Fontanesi, A. Micheli, M. Podda, D. Tortorella (2026). "A method for the systematic generation of graph XAI benchmarks via Weisfeiler–Leman coloring." Data Mining and Knowledge Discovery, vol. 40(4), 42.
Read Paper
Published in Neurocomputing, 2026
RIM4G is a new model for graph node representation that is ameanable to physical implementation on non-conventional hardware.
Recommended citation: M.G. Berni, A. Brau, A. Micheli, D. Tortorella (2026). "Randomized Ising models for graph node representation." Neurocomputing, vol. 679, 133195.
Read Paper
Published in Machine Learning, 2025
We propose XNQ, an efficient model for network quantification that scales on large networks and is resilient to heterophily.
Recommended citation: A. Micheli, A. Moreo, M. Podda, F. Sebastiani, W. Simoni, D. Tortorella (2025). "Efficient quantification on large-scale networks." Machine Learning, vol. 114, 270.
Read Paper
Published in Machine Learning, 2025
We apply graph signal processing to node importances of XAI explanations to refine the characterization of DGN inductive biases.
Recommended citation: M. Fontanesi, A. Micheli, M. Podda, D. Tortorella (2025). "Bridging XAI and spectral analysis to investigate the inductive biases of deep graph networks." Machine Learning, vol. 114, 257.
Read Paper
Published in Pattern Recognition Letters, 2025
We propse an evaluation framework for graph rewiring methods based on training-free GNNs. In our fair conditions, rewiring rarely improves message-passing on 12 node and graph classification tasks.
Recommended citation: A. Micheli, D. Tortorella (2025). "An Empirical Evaluation of Rewiring Approaches in Graph Neural Networks." Pattern Recognition Letters, vol. 196, pp. 134–141.
Read Paper
Published in Neurocomputing, 2024
Analysis of dense and sparse reservoir designs for node-level GESN via topology-dependent and topology-agnostic richness measures for node embeddings.
Recommended citation: A. Micheli, D. Tortorella (2024). "Designs of Graph Echo State Networks for Node Classification." Neurocomputing, vol. 597, 127965.
Read Paper
Published in Neurocomputing, 2023
Node-level GESN is highly effective for heterophilic node classification tasks, while also being efficient and resilient to over-smoothing.
Recommended citation: A. Micheli, D. Tortorella (2023). "Addressing Heterophily in Node Classification with Graph Echo State Networks." Neurocomputing, vol. 550, 126506.
Read Paper
Published in Neurocomputing, 2022
DynGESN is introduced as a novel reservoir computing model for temporal graphs. More efficient then temporal graph kernels and 100x faster than temporal GNNs.
Recommended citation: A. Micheli, D. Tortorella (2022). "Discrete-Time Dynamic Graph Echo State Networks." Neurocomputing, vol. 496, pp. 85-95.
Read Paper
Published in ICANN, 2025
An analysis of time-scales in Deep Bidirectional ESNs applied to sentence-level and token-level NLP tasks.
Recommended citation: C. Baccheschi, A. Bondielli, A. Lenci, A. Micheli, L. Passaro, M. Podda, D. Tortorella (2025). "Investigating Time-Scales in Deep Echo State Networks for Natural Language Processing." Artificial Neural Networks and Machine Learning. ICANN 2025 International Workshops and Special Sessions, LNCS vol. 16072, pp. 188-200.
Read Paper
Published in GbRPR, 2025
We investigate shortcut learning in graphs by experimenting with adding training counterfactuals and shifting inductive bias.
Recommended citation: D. Tortorella, M. Fontanesi, A. Micheli, M. Podda (2025). "An Empirical Investigation of Shortcuts in Graph Learning." Graph-Based Representations in Pattern Recognition. GbRPR 2025. LNCS vol. 15727, pp. 147-156.
Read Paper
Published in ESANN, 2025
Reframing the prediction of dynamical properties of biochemical pathways as a link prediction task over the protein-protein interaction networks (PPINs).
Recommended citation: A. Dipalma, D. Tortorella, A. Micheli (2025). "Robustness in Protein-Protein Interaction Networks: A Link Prediction Approach." Proceedings of the 33rd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2025), pp. 277-282.
Read Paper
Published in ESANN, 2025
We propose a RIM, a new model for encoding graph topology that is ameanable to physical implementation on non-conventional hardware.
Recommended citation: D. Tortorella, A. Brau, A. Micheli (2025). "Encoding Graph Topology with Randomized Ising Models." Proceedings of the 33rd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2025), pp. 271-276.
Read Paper
Published in Discovery Science, 2024
We analyze the alignment of DGNs explanations to node centrality and graph connectivity, highlighting the presence of different inductive biases.
Recommended citation: M. Fontanesi, A. Micheli, M. Podda, D. Tortorella (2025). "Analyzing Explanations of Deep Graph Networks through Node Centrality and Connectivity." Proceedings of the 27th International Conference on Discovery Science (DS 2024), LNCS vol. 15243, pp. 295-309.
Read Paper
Published in ESANN, 2024
GESN relieves part of catastrophic forgetting in the continual learning setting by avoiding training representations for graph classification.
Recommended citation: D. Tortorella, A. Micheli (2024). "Continual Learning with Graph Reservoirs: Preliminary experiments in graph classification." Proceedings of the 32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2024), pp. 35-40.
Read Paper
Published in ICANN, 2024
A preliminary analysis of dynamical properties of Onion ESN, a novel reservoir with groups of units presentig an annular spectrum.
Recommended citation: D. Tortorella, A. Micheli (2024). "Onion Echo State Networks: A Preliminary Analysis of Dynamics." Proceedings of the 33rd International Conference on Artificial Neural Networks (ICANN 2024), LNCS vol. 15025, pp. 117-128.
Read Paper
Published in ECML PKDD, 2024
A continuous-depth ESN is proposed, where a smooth depth hyperparameter regulates the extent of local connections.
Recommended citation: A. Ceni, P. F. Dominey, C. Gallicchio, A. Micheli, L. Pedrelli, D. Tortorella (2024). "Continuously Deep Recurrent Neural Networks." Machine Learning and Knowledge Discovery in Databases: Research Track. ECML PKDD 2024. LNCS vol. 14947, pp. 59-73.
Read Paper
Published in ESANN, 2023
Adding a representation entropy term into the loss of Hinton’s FFA improves accuracy.
Recommended citation: M. Pardi, D. Tortorella, A. Micheli (2023). "Entropy Based Regularization Improves Performance in the Forward-Forward Algorithm." Proceedings of the 31st European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2023), pp. 393-398.
Read Paper
Published in ESANN, 2023
Preliminary analysis of GESN’s node embedding richness via entropy and numerical analysis metrics.
Recommended citation: D. Tortorella, A. Micheli (2023). "Richness of Node Embeddings in Graph Echo State Networks." Proceedings of the 31st European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2023), pp. 11-16.
Read Paper
Published in GbRPR, 2023
Minimizing the number of support vectors in SVM without any loss of accuracy via an RRQR factorization of kernel matrix.
Recommended citation: D. Tortorella, A. Micheli (2023). "Minimum Spanning Set Selection in Graph Kernels." Graph-Based Representations in Pattern Recognition. GbRPR 2023. LNCS vol. 14121, pp. 15-24.
Read Paper
Published in LoG, 2022
GESN achieves a significantly better accuracy on six heterophilic node classification tasks via tuning Lipschitz constants instead of resorting to graph rewiring.
Recommended citation: D. Tortorella, A. Micheli (2022). "Leave Graphs Alone: Addressing Over-Squashing without Rewiring" (Extended Abstract). Presented at the First Learning on Graphs Conference (LoG 2022), Virtual Event, December 9–12, 2022.
Read Paper
Published in ESANN, 2022
Preliminary experiments on heterophilic node classification with GESN, showing the effectiveness of going beyond stability constraints.
Recommended citation: D. Tortorella, A. Micheli (2022). "Beyond Homophily with Graph Echo State Networks." Proceedings of the 30th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2022), pp. 491-496.
Read Paper
Published in ICANN, 2022
An in-depth theoretical analysis of asymptotic dynamics in Deep ESNs with different contractivity hierarchies.
Recommended citation: D. Tortorella, C. Gallicchio, A. Micheli (2022). "Hierarchical Dynamics in Deep Echo State Networks." Proceedings of the 31st International Conference on Artificial Neural Networks (ICANN 2022), pp. 668-679.
Read Paper
Published in IJCNN, 2022
More accurate stability bounds for GESN based on graph spectral properties.
Recommended citation: D. Tortorella, C. Gallicchio, A. Micheli (2022). "Spectral Bounds for Graph Echo State Network Stability." Proceedings of the 2022 International Joint Conference on Neural Networks.
Read Paper
Published in ESANN, 2021
Preliminary experiments on temporal graph classification with DynGESN, a novel reservoir computing model for dynamic graphs.
Recommended citation: D. Tortorella, A. Micheli (2021). "Dynamic Graph Echo State Networks." Proceedings of the 29th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2021), pp. 99-104.
Read Paper