analiza aiqbench
This paper addresses the critical transparency gap in high-capacity neural networks by proposing NeuroRule, a framework that distills black-box models into concise, explainable propositional logic rules. By adapting evolutionary rule-set infrastructure, the approach successfully maintains model performance while eliminating the need for access to original training data, thereby enhancing trustworthiness for real-world deployment.
NeuroRule: Making Black-Box Neural Networks Explainable through Rule-set Evolution
arXiv:2609.26841v1 Announce Type: cross Abstract: High-capacity neural network models have achieved state-of-the-art performance across diverse classification tasks, yet they frequently operate as black-box models, lacking the transparency necessary for critical decision-making. Such opacity creates a persistent trade-off between performance and explainability. This paper proposes a solution to address this gap: the NeuroRule knowledge distillation framework that results in explainable rule-sets from neural network models. NeuroRule adapts the EVOTER rule-set evolution infrastructure to treat neural networks as targets for the evolution process, distilling their performance into concise sets of propositional logic expressions. There are three primary contributions: (1) an evolutionary method for distilling black-box neural network models into explicit rule-set models; (2) a method for making rule sets more explainable by including a conciseness objective to evolution; and (3) a demonstration that the distillation is viable even without access to the original neural network training data. The paper thus establishes that black-box neural network models can be made explainable and therefore useful in real-world applications where trustworthiness is paramount.
Abstract:High-capacity neural network models have achieved state-of-the-art performance across diverse classification tasks, yet they frequently operate as black-box models, lacking the transparency necessary for critical decision-making. Such opacity creates a persistent trade-off between performance and explainability. This paper proposes a solution to address this gap: the NeuroRule knowledge distillation framework that results in explainable rule-sets from neural network models. NeuroRule adapts the EVOTER rule-set evolution infrastructure to treat neural networks as targets for the evolution process, distilling their performance into concise sets of propositional logic expressions. There are three primary contributions: (1) an evolutionary method for distilling black-box neural network models into explicit rule-set models; (2) a method for making rule sets more explainable by including a conciseness objective to evolution; and (3) a demonstration that the distillation is viable even without access to the original neural network training data. The paper thus establishes that black-box neural network models can be made explainable and therefore useful in real-world applications where trustworthiness is paramount.
Neural and Evolutionary Computing (cs.NE); Machine Learning (cs.LG)
arXiv:2609.26841 [cs.NE]
(or arXiv:2609.26841v1 [cs.NE] for this version)
https://doi.org/10.48550/arXiv.2609.26841 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 260 cuvinte
Subiecte conexe
Dialogue with Eliezer Yudkowsky on Neural Networks
LessWrong · 2 oct. 2026
From Black Box to Executable Logic: Explainable Reinforcement Learning through Prolog Expert Systems
arXiv · 21 iul. 2026
From Black Box to Executable Logic: Explainable Reinforcement Learning through Prolog Expert Systems
arXiv · 20 iul. 2026
Humanity’s Last Invention — Richard Socher of Recursive
Latent Space · 14 sept. 2026