Explanatory predictions with artificial neural networks and argumentation
Oana Cocarascu, Kristijonas Čyras, Francesca Toni · Spiral (Imperial College London) · 2018
Data-centric AI has proven successful in several domains, but its outputs are often hard to explain. We present an architecture combining Artificial Neural Networks (ANNs) for feature selection and an instance of Abstract Argumentation (AA) for reasoning to provide effective predictions, explain- able both dialectically and logically. In particular, we train an autoencoder to rank features in input ex- amples, and select highest-ranked features to gen- erate an AA framework that can be used for mak- ing and explaining predictions as well as mapped onto logical rules, which can equivalently be used for making predictions and for explaining. We show empirically that our method significantly out- performs ANNs and a decision-tree-based method from which logical rules can also be extracted.