Value-based Argumentation Frameworks as Neural-symbolic Learning Systems
Journal of Logic and Computation
This is the unspecified version of the paper. This version of the publication may differ from the final published version. Permanent repository link: http://openaccess.city.ac.uk/294/ Link to published version: http://dx. Abstract While neural networks have been successfully used in a number of machine learning applications, logical languages have been the standard for the representation of argumentative reasoning. In this paper, we establish a relationship between neural networks and
... works and argumentation networks, combining reasoning and learning in the same argumentation framework. We do so by presenting a new neural argumentation algorithm, responsible for translating argumentation networks into standard neural networks. We then show a correspondence between the two networks. The algorithm works not only for acyclic argumentation networks, but also for circular networks, and it enables the accrual of arguments through learning as well as the parallel computation of arguments. has been advocated by Valiant as a key challenge for computer science  . In this paper, we introduce a neural argumentation algorithm, which is responsible for translating value-based argumentation networks into standard neural networks with the use of neural-symbolic systems  . Neural-symbolic systems concern the application of problem-specific symbolic knowledge within the neurocomputing paradigm. They have been used to combine neural network-based learning systems with nonmonotonic, epistemic, and temporal symbolic knowledge representation and reasoning [13, 14] . We show that the neural network created by the neural argumentation algorithm executes a sound computation of the prevailing arguments in the argumentation network. This shows that the two representations are equivalent. However, arguments will frequently attack one another in such a way that cycles are formed. In such cases, a notion of relative strength of the arguments may be required to decide which arguments should prevail. Still, in some cases, circularities may lead to an infinite loop in the computation. To tackle this problem, we propose the use of a learning mechanism. Learning can be used to resolve circularities by the iterative change of the strength of arguments as new information becomes available. Learning and its relation to accrual, cumulative argumentation [32, 27] in neural networks will also be discussed. Our long-term goal is to facilitate learning capabilities in value-based argumentation frameworks, as arguments may evolve over time, with certain arguments being strengthened and others weakened. At the same time, we seek to enable the parallel computation of argumentation frameworks by making use of the machinery of neural networks. The remainder of the paper is organised as follows. In Section 2, we present the basic concepts of value-based argumentation, neural networks, and neural-symbolic systems used throughout the paper. In Section 3, we introduce the neural argumentation algorithm, and prove that the neural network executes a sound computation of the argumentation network, and therefore that the translation is correct. In Section 4, we investigate how learning may help overcome circularities, and how neural networks may support accrual, cumulative argumentation. Section 5 concludes the paper and discusses directions for future work.