摘要
In the research of rule extraction from neural networks, fidelity describeshow well the rules mimic the behavior of a neural network while accuracy describes how well therules can be generalized. This paper identifies the fidelity-acuracy dilemma. It argues todistinguish rule extraction using neural networks and rule extraction for neural networks accordingto their different goals, where fidelity and accuracy should be excluded from the rule qualityevaluation framework, respectively.
In the research of rule extraction from neural networks, fidelity describeshow well the rules mimic the behavior of a neural network while accuracy describes how well therules can be generalized. This paper identifies the fidelity-acuracy dilemma. It argues todistinguish rule extraction using neural networks and rule extraction for neural networks accordingto their different goals, where fidelity and accuracy should be excluded from the rule qualityevaluation framework, respectively.
基金
国家杰出青年科学基金，国家自然科学基金
