@inproceedings{67339,
  abstract     = {{Explainable artificial intelligence (XAI) is essential for critical domains such as healthcare and autonomous systems to build trust and confidence in real-world deployment. In this context, description logic knowledge bases (KBs) provide structured and semantically rich representations that support reasoning and informed decision-making. A core task in applying KBs to XAI is class expression learning (CEL), which generates explainable logical descriptions for classifying instances within KBs. Unlike black-box models with opaque internal mechanisms, CEL provides global explainability and ease of integration with domain knowledge. However, current approaches to CEL face significant limitations such as poor scalability, failure to capture rare patterns, and limited exploration of the vast class expression search space. To overcome these limitations, we introduce LYRA, a novel multi-agent deep reinforcement learning framework that formulates CEL as a collaborative planning task under uncertainty. The integration of the Dempster–Shafer theory enables agents to effectively reason under ambiguity and manage conflicting or inconsistent information. Our experiments show that LYRA outperforms state-of-the-art methods on seven out of eight datasets, demonstrating robust and scalable CEL. Additionally, LYRA offers interpretable decisions and employs advanced search strategies, enabling the discovery of more precise and expressive class expressions than existing approaches.}},
  author       = {{Abdulmaqsod, Amgad and Mahmood, Yasir and Ngonga Ngomo, Axel-Cyrille and Sherif, Mohamed Ahmed}},
  booktitle    = {{The Semantic Web – ISWC 2026}},
  keywords     = {{amgad dice enexa fairomics mahmood ngonga sailproject sherif simba whale}},
  title        = {{{LYRA: Belief-Driven Scalable Class Expression Learning in Description Logics}}},
  year         = {{2026}},
}