This special track explores approaches to Explainable AI (XAI) based on logic and reasoning techniques. While the traditional role of logic in AI lies in knowledge representation and reasoning, it is increasingly used in other areas highly relevant to Explainable AI. This includes neuro-symbolic approaches to AI, learning from a combination of knowledge and data, explanation interfaces, and as a basis for meta-reasoning about machine learning systems. An example of meta-reasoning is explaining predictions made by classifiers. As opposed to popular heuristic-based approaches, the logic-based approach, based on prime implicants, leads to provably correct explanations and the ability to reason formally about the behaviour of a classifier, such as whether the classifier exhibits bias. Logic and reasoning can be used to study the foundations of explanations in diverse settings. This track aims to bring together researchers applying logic and reasoning within XAI. It covers various logic-based strategies for representing, validating, and implementing XAI systems, emphasising the harmonisation of logic with AI explainability.
Topics
| Abduction-based explanations for machine learning models |
| Completeness investigation of reasons behind decisions |
| Logical framework for xAI in classification systems |
| Non-monotonic background theories for boolean classifiers explanation |
| Logic-based explanations for neural networks |
| Logic-based reasoning techniques for classifier bias |
| Uncertainty quantifiers for explainability |
| Logic and reasoning for explainable neuro-symbolic AI |
| Hard logical constraints for multi-label supervised neural networks |
| Explainability for machine reasoning techniques |
| Enswer set programming-based explanation approaches |
| XAI methods for inconsistencies handling/discovering |
| Explainability in theorem proving |
| End-to-end differentiable proving for xAI |
| Symbolic/sub-symbolic techniques integration for XAI |
| Interface development for logic-based explanations |
| Investigation of the impact of logic-style explanations on user decisions |
| Logic-style explanations under different levels of uncertainty |
| Logical foundations of explanations |
| Axiomatic foundations of explainability |
| Logic-based validation of heuristic for XAI |
| Explainable normative supervisors for reinforcement learning agents |
| Explanations for deontic logic |
Supported by
