Learning Quantitative Argumentation Frameworks Using Sparse Neural Networks and Swarm Intelligence Algorithms

Quantitative Argumentation Frameworks represent arguments and the relationships among them, such as support and attack relationships, in a graph structure. Even though this is not their main purpose, they will be used to solve classification problems by following a new approach. This approach is based on constructing them out of sparse multilayer perceptrons (MLP) and exploiting their advantage of being easily interpreted. In this thesis, I will develop learning algorithms for argumentative classifiers will be developed and their performance will be evaluated. To do so, sparse graph structures will be generated, that can be seen as MLPs, and their parameters can be learned using the usual backpropagation procedure. Finally, the results of this approach will be evaluated. We hope to reach findings that will prove this method as an interesting explainable machine learning tool and to create a neural network that is not a total black box.