abductive reasoning

Existing explainable AI (XAI) methods lack probabilistic explanation mechanisms that jointly ensure transparency and user adaptability under uncertainty. Method: We propose the first unified framework generating two complementary types of probabilistic explanations: (1) self-contained, user-agnostic singleton explanations; and (2) model-harmonized explanations grounded in users’ probabilistic mental models. Our approach innovatively extends model harmonization to probabilistic human models, introduces the novel quantitative metrics “explanation gain” and “explanatory power”, and leverages MUS/MCS duality for efficient computation. The framework integrates probabilistic logic programming, Bayesian knowledge representation, and logic-based harmonization techniques. Results: Extensive experiments across diverse benchmarks demonstrate significant improvements in explanation credibility, user adaptability, and computational efficiency—validating both the effectiveness and scalability of the framework.

This study addresses the critical challenge of balancing provable validity and fusion efficiency when integrating multi-source information within inferential models (IMs). Focusing on possibility-measure-based IMs, the work proposes a general validity-preserving fusion framework applicable across diverse dependence structures—including independence, arbitrary dependence, and exchangeability. By employing a rank-and-calibrate construction, the framework achieves robust fusion while rigorously maintaining the theoretical validity guarantees inherent to IMs. The research establishes, for the first time, a universal mechanism for preserving validity under fusion, exposes the inefficiency of conventional fusion operators in the IM context, and offers superior alternatives that significantly enhance fusion efficiency without compromising statistical rigor.

To address declining model interpretability and consequent trust deficits in complex machine learning systems, this paper proposes a trustworthy XAI framework grounded in uncertainty decoupling. Methodologically, it is the first to disentangle aleatoric from epistemic uncertainty and leverages epistemic uncertainty as a reliability metric for explanations: (i) as a rejection threshold for low-fidelity explanations, and (ii) as a dynamic signal to adapt explanation strategies—e.g., feature attribution or counterfactual generation—to local uncertainty conditions. The framework integrates Bayesian neural networks, rigorous uncertainty quantification, and state-of-the-art XAI techniques. Extensive experiments across diverse models—including traditional machine learning and deep neural networks—demonstrate substantial improvements in explanation stability and robustness. Crucially, it effectively filters unreliable explanations, thereby enhancing user trust in AI-driven decisions.

This work addresses the limitations of existing counterfactual explanation methods, which are often confined to local instances and a single explanation type, lacking a global perspective and systematic categorization. The authors propose an axiomatic framework that characterizes the capability boundaries of counterfactual explainers through formally defined desiderata and establish impossibility theorems demonstrating that certain combinations of these axioms cannot be simultaneously satisfied. Building on this foundation, they construct a one-to-one correspondence between five compatible subsets of axioms and five fundamentally distinct types of counterfactual explanations. By leveraging representation theorems, formal logic, and complexity analysis, the study provides the first unified classification of mainstream counterfactual explainers, clarifying their behavioral properties and computational complexities.

This paper addresses the pervasive “explanation inconsistency” problem in post-hoc interpretability—where distinct explanation methods assign contradictory feature attributions to the same model prediction. We formalize explanation divergence, then conduct a systematic empirical investigation involving in-depth interviews with 24 data scientists, extensive evaluation across 4 datasets, 6 models, and 6 explanation methods, and controlled online user studies. Our findings reveal that explanation divergence occurs frequently and lacks principled theoretical criteria for resolution. We introduce the first quantitative framework for measuring and characterizing such divergence. Crucially, we find practitioners rely heavily on ad hoc heuristics—e.g., favoring local versus global methods or visually intuitive outputs—yet these strategies exhibit low reliability and poor reproducibility, undermining decision trustworthiness in high-stakes applications. Our work establishes foundational insights for robust interpretability evaluation and paves the way for principled, synergistic explanation paradigms.

This study addresses the question of whether large language models (LLMs) update their probabilistic beliefs in a manner consistent with Bayesian principles when faced with new evidence—a critical capability for reasoning under uncertainty. The authors introduce, for the first time, a metric termed the “information processing gap” to systematically evaluate belief-updating consistency in mainstream LLMs. Through comprehensive analyses involving multi-source evidence integration, comparison against normative Bayesian inference, heuristic behavior characterization, and consistency diagnostics, the work reveals that LLMs predominantly rely on non-Bayesian heuristic strategies. Notably, these heuristics often outperform strict Bayesian updating in downstream tasks, suggesting a fundamental misspecification in the models’ internal world representations. The study further provides an interpretable diagnostic framework to characterize such deviations from rational belief updating.

This work addresses a key limitation in existing multi-agent belief fusion approaches, which typically assume a fixed cognitive partition structure and thus struggle to accommodate dynamic adjustments of the representation space caused by runtime changes in observational capabilities. The paper proposes a formal framework that enables dynamic evolution of cognitive partitions under continuous belief profiles, achieving for the first time interpretable belief reconciliation: refinement preserves admissibility, while coarsening ensures consistency through a unique mass-conserving repair mechanism, accompanied by complete explanations. The framework integrates the declarative constraint reasoning of Answer Set Programming (ASP) with Python’s numerical computation capabilities, leveraging ASP to manage structural changes and generate explanations. Empirical evaluation across 100 randomly generated topological change scenarios demonstrates 100% detection of constraint violations and full explanation coverage, confirming the method’s effectiveness and completeness.

Existing methods for logical explanations often struggle to simultaneously ensure correctness and achieve broad coverage. To address this challenge, this work proposes two novel approaches—Onestep and Twostep—that generate high-coverage logical-form explanations while strictly preserving correctness. Onestep employs a single-step optimization framework, whereas Twostep adopts a progressive strategy that iteratively refines explanations. Both methods integrate logical rule generation, boundary optimization, feature bound adjustment, and coverage maximization techniques. Experimental results demonstrate that Twostep significantly outperforms current state-of-the-art methods, improving average explanation coverage by 72.60% across multiple datasets.

This work addresses the instability and high computational cost of existing post-hoc explanation methods in regions near ambiguous decision boundaries, which undermine explanation reliability. To tackle this challenge, the study proposes leveraging epistemic uncertainty—estimated via Bayesian deep learning—as a low-cost proxy for explanation reliability, dynamically determining when to generate explanations to optimize the trade-off between quality and computational cost under resource constraints. The approach is systematically evaluated across multiple XAI techniques (e.g., LIME, SHAP) and model architectures, revealing a strong negative correlation between epistemic uncertainty and explanation stability. Extensive experiments on four tabular datasets, five model types, and four XAI methods validate this relationship, while additional image classification tasks demonstrate the method’s generalizability.

This work addresses the redundancy commonly observed in unsupervised anomaly detection ensembles, where constituent detectors often rely on similar decision rationales, limiting their complementarity. To overcome this, the study introduces, for the first time, the use of SHAP explanations to quantify each detector’s feature importance assignments, thereby constructing an “explanation profile” to measure inter-model similarity. Based on this, a novel model selection criterion grounded in explanation diversity is proposed. Both theoretical analysis and empirical evaluation demonstrate a strong correlation between explanation dissimilarity and detection complementarity. Ensembles formed by combining high-performing individual detectors with diverse explanation profiles consistently outperform state-of-the-art methods across multiple benchmark datasets, confirming the effectiveness and novelty of the proposed strategy.