The article proposes a way to incorporate expert rules into machine learning models, specifically in the context of concept-based learning. The authors discuss methods to combine logical rules and neural networks to predict concept probabilities. They also present approaches to train neural networks to ensure the output probabilities of concepts adhere to expert rules. This solution can be seen as a blend of inductive and deductive learning. The use of expert rules is broad, considering any logical function that connects concepts and class labels or just concepts with each other as a rule. This expands the scope of the proposed results significantly.
Publication date: 22 Feb 2024
Project Page: https://arxiv.org/abs/2402.14726
Paper: https://arxiv.org/pdf/2402.14726
