神经网络如何学习自身的欺诈规则：一种神经符号人工智能实验

Most neuro-symbolic systems rely on human-written rules, but this experiment explores a novel approach where a neural network can autonomously discover rules. By extending a hybrid neural network with a differentiable rule-learning module, the model automatically extracts IF-THEN fraud rules during training. On the Kaggle Credit Card Fraud dataset, which has a fraud rate of 0.17%, the model successfully learned interpretable rules. This experiment demonstrates the potential of neural networks in fraud detection, particularly in the areas of rule generation and interpretability.