grounded rule generation

Existing methods for generating detection rules rely on specific input-output pairs and lack a unified framework. This work formalizes the task for the first time as a unified mapping from contextual and target-language inputs to detection rules, introducing UniRule—a novel framework that models semantic distance in a dual semantic projection space encompassing detection intent and detection logic to characterize optimal rules. UniRule integrates an agent-based RAG architecture to achieve generalization across diverse contexts and languages. Experimental results demonstrate that UniRule significantly outperforms pure large language model (LLM) approaches across twelve distinct scenarios, achieving a Bradley-Terry preference coefficient of 0.52, thereby validating its effectiveness and broad applicability.

This paper addresses the challenge of transforming tacit human expertise into computable, executable, and interpretable knowledge. To this end, it proposes a knowledge representation and reasoning framework tailored for dynamic, complex scenarios. Methodologically, it introduces the novel “Knowledge Cloud” model, integrating dynamic relational modeling, explicit reasoning-path characterization, and cloud-refinement mechanisms—grounded in Minsky’s frame theory, the KSYNTH knowledge description language, a General Paradigm Pattern Builder (GPPB), and eXplainable AI (XAI) design principles. Unlike static ontologies, rule-based systems, or conventional knowledge graphs, the framework enables adaptive knowledge evolution and traceable, auditable inference. Empirical validation across three distinct domains—naval combat simulation, water treatment fault diagnosis, and RISK strategic decision-making—demonstrates substantial improvements in knowledge expressivity, task adaptability, and decision interpretability.

This study addresses the challenge of aligning machine reasoning with human cognition, focusing on synergistic modeling between symbolic and parametric knowledge bases. Method: We propose a unified analytical framework centered on knowledge base typology, introducing the first formal “symbolic–parametric” dichotomy theory and identifying hybrid reasoning as the critical pathway toward human–machine intelligence alignment. Integrating knowledge representation and reasoning (KRR), symbolic logic, neurosymbolic computing, implicit knowledge extraction from large language models, and knowledge fusion architectures, we construct a comprehensive methodology map for knowledge-base-driven reasoning across the full spectrum of paradigms. Contribution/Results: The work identifies three fundamental bottlenecks—interpretability, generalization, and dynamic knowledge updating—and establishes neurosymbolic integration as the pivotal direction for overcoming them, thereby advancing principled, aligned, and adaptive AI reasoning systems.

Existing rule-based knowledge graph completion methods offer interpretability but rely on large rule sets, compromising both interpretability and efficiency. This paper introduces the concept of *rule context*—a cohesive subset of collaboratively functioning rules modeled as a probabilistic circuit—enabling traceable probabilistic inference without independence assumptions while preserving logical semantics. By learning the probability distribution over rule contexts, our approach drastically compresses the rule set while supporting both exact and approximate query probability estimation. Evaluated on eight benchmark datasets, our method achieves performance comparable to state-of-the-art baselines (e.g., AnyBURL) using only 4–30% of their rules; with merely 4% of the rules, it retains 91% of peak accuracy. Moreover, inference speed improves by up to 31×. The framework thus unifies high efficiency, faithful interpretability, and rigorous probabilistic semantics.

This work addresses critical challenges in safety-critical rule-based systems—namely poor scalability, fragility, and goal mis-specification—which often lead to reward hacking and failures in formal verification. To overcome these limitations, the authors propose a neuro-symbolic causal framework that integrates first-order logic abductive trees, structural causal models, and deep reinforcement learning within a MAPE-K control loop. A novel meta-layer architecture enables the automatic synthesis and formal verification of rules from natural language objectives. This meta-layer comprises a goal/rule synthesizer and a rule verification engine, which iteratively generate necessary and sufficient causal rule sets grounded in legal and safety principles provided by human experts. Evaluated in an autonomous driving scenario, the approach successfully derives a minimal yet complete rule set, formally encoded as logical constraints, demonstrating its modularity, traceability, and practical applicability.

This work addresses the challenge that large language models often struggle to effectively leverage feedback during mathematical reasoning due to context inflation and the conflation of knowledge storage with usage. To resolve this, the authors propose a decoupled framework featuring a hierarchical knowledge base organized by problem difficulty and domain, coupled with a dynamic reasoning mechanism. This mechanism employs self-reflective learning to construct cognitive trees and integrates tiered self-consistency with gated retrieval to dynamically match problem difficulty and retrieve relevant knowledge branches as needed. Evaluated on AIME 2025, the approach achieves 62.2% accuracy—surpassing the fixed Best-of-5 baseline by 10.4 percentage points and outperforming the strongest existing method, Tiered + GEPA, by 5.6 points—while also delivering consistent gains on the MATH-HARD and OlymMATH subsets.

This work proposes a context-mediated domain adaptation approach that treats expert edits of AI-generated content not merely as endpoint corrections but as implicit carriers of domain knowledge. By inversely analyzing user editing behaviors on multi-agent generated outputs, the method dynamically distills implicit norms to guide large language model–driven reasoning. It establishes a bidirectional semantic link between generated content and system inference, enabling norm bootstrapping, knowledge transfer, and in-context learning—all grounded in edit patterns as a novel form of implicit knowledge representation. Integrated into the web framework Seedentia, the approach successfully extracted 46 domain-specific rules from expert revisions, demonstrating the efficacy of this paradigm in capturing and leveraging tacit expertise for adaptive AI systems.

This study addresses the frequent inefficiencies in human-AI collaboration caused by incomplete contextual information, which often leads to excessive iteration and suboptimal output quality. To mitigate this, the authors propose a structured context construction framework that integrates a five-role context package—comprising authority, exemplars, constraints, evaluation criteria, and metadata—within a four-stage workflow encompassing review, design, construction, and audit. Notably, this work pioneers the incorporation of information theory and reliability engineering principles into context quality assessment, yielding a reusable and auditable collaboration framework. Empirical results from 200 interaction trials demonstrate that the approach reduces the average number of iterations from 3.8 to 2.0, increases first-pass success rates from 32% to 55%, and achieves a final task success rate of 91.5%.

Existing Hybrid MKNF knowledge bases lack support for classical negation in their rule component, making it difficult to express explicit negative knowledge and thereby limiting their applicability in safety-critical scenarios. This work addresses this limitation by formally integrating classical negation into the Hybrid MKNF rule component through a principled fusion of description logics and logic programming. The paper presents an extended Hybrid MKNF language that accommodates classical negation, rigorously defines its syntax and semantics, and introduces a novel inference framework grounded in well-founded semantics. A key contribution is the development of a general and efficient procedure for computing well-founded models of the extended language, substantially enhancing the system’s capacity to represent and reason with explicit negative information.