Equivariant Transition Matrices for Explainable Deep Learning: A Lie Group Linearization Approach
Machine Learning and Knowledge Extraction, Vol. 8, No. 4, Article 92
Editorial summary
The study makes explanations of a trained neural network more consistent when inputs are transformed, for example by rotation. It adds Lie-group constraints to transition matrices after training. Tests on synthetic data and handwritten digits compare this consistency with how closely the explanation follows the model.
Bibliographic reference
Use the publisher’s record for the citation format required by your journal or organisation.
Pavlo Radiuk, Oleksander Barmak, Leonid Bedratyuk, Iurii Krak. "Equivariant Transition Matrices for Explainable Deep Learning: A Lie Group Linearization Approach". Machine Learning and Knowledge Extraction, Vol. 8, No. 4, Article 92, 2026. https://doi.org/10.3390/make8040092
Test a method against your business task.
Describe the decision your system needs to support and the data you have. We assess whether a method fits, define a prototype and agree how to measure its quality, response time and operating cost.