THE TRUTH WILL COME TO LIGHT - DIRECTIONS AND CHALLENGES IN EXTRACTING THE KNOWLEDGE EMBEDDED WITHIN TRAINED ARTIFICIAL NEURAL NETWORKS

To date, the preponderance of techniques for eliciting the knowledge e mbedded in trained artificial neural networks (ANN's) has focused prim arily on extracting rule-based explanations from feedforward ANN's, Th e ADT taxonomy for categorizing suck techniques was proposed in 1995 t o provide a basis for the systematic comparison of the different appro aches. This paper shows that not only is this taxonomy applicable to a cross section of current techniques for extracting rules from trained feedforward ANN's but also how the taxonomy can be adapted and extend ed to embrace a broader range of ANN types (e.g., recurrent neural net works) and explanation structures. In addition the paper identifies so me of the key research questions in extracting the knowledge embedded within ANN's including the need for the formulation of a consistent th eoretical basis for what has been, until recently, a disparate collect ion of empirical results.