This paper presents an extension to the EVOTER (Evolution of Transparent Explainable Rule Sets) framework, previously introduced to promote transparency and explainability in AI through the evolution of interpretable rule sets. While EVOTER demonstrated strong performance in generating compact and understandable rule-based models using a list-based grammar, the current work, Accelerated Evolutionary Rule-based Learning (AERL), builds upon this foundation by addressing scalability challenges. Specifically, AERL leverages GPU-accelerated computation and back-propagation fine-tuning within a PyTorch framework to significantly enhance the efficiency of rule evaluation and evolution. By introducing a tensorized numerical representation of rules and incorporating gradient-based optimization, AERL achieves superior computational speed and improved search-space exploration. Experimental results confirm that the approach is effective, achieving faster convergence and similar accuracy. It thus offers a scalable path forward for explainable AI.

PDF

In Proceedings of the Genetic and Evolutionary Computation Conference Companion: Workshop on Evolutionary Rule-based Machine Learning, 2025.

@inproceedings{shahrzad:geccows25, title={GPU-Accelerated Rule Evaluation and Evolution}, author={Hormoz Shahrzad and Risto Miikkulainen}, booktitle={Proceedings of the Genetic and Evolutionary Computation Conference Companion: Workshop on Evolutionary Rule-based Machine Learning}, month={ }, url="http://nn.cs.utexas.edu/?shahrzad:geccows25", year={2025} }