Domain Project

We are proud to announce that the paper “Learning Brave Assumption-Based Argumentation Frameworks via ASP” has been accepted at the 27th European Conference on Artificial Intelligence (ECAI 2024).

This work addresses a key challenge in non-monotonic reasoning: how to automatically learn assumption-based argumentation (ABA) frameworks from examples. While traditional approaches assume the ABA structure is provided in advance, this paper introduces a method to construct such frameworks directly from data, based on brave reasoning under stable extensions—a setting where a statement is accepted if it is supported in at least one extension.

We propose a novel transformation-based learning strategy, implemented via Answer Set Programming (ASP). The resulting system, ASP-ABAlearnB, applies a series of rewriting and generalization steps—including assumption introduction and rule folding—to generate ABA frameworks that fit both positive and negative examples. The approach guarantees termination and correctness under specific conditions and supports the derivation of generalized, intensional rules.

Empirical evaluation on established benchmarks—including variants of the Nixon Diamond and real-world datasets—shows that the proposed method achieves competitive or superior performance compared to state-of-the-art inductive logic programming systems such as ILASP, particularly in scenarios where learning from defeasible or incomplete knowledge is required.

This work highlights the power of combining argumentation theory with logic-based learning, and we look forward to presenting it to the AI community at ECAI 2024.