conflict resolution

Evaluating decision consistency among experts (or classifiers) in unsupervised settings—without ground-truth labels or prior knowledge—remains challenging. Method: We propose a logic-based evaluation framework that models expert scores as an integer linear programming problem, integrating statistical alignment signals with formal logical constraints—including inequality constraints and a novel, universally applicable linear equality axiom for arbitrary finite test sets. This axiom ensures logical satisfiability of collective judgments. Contribution/Results: Our framework introduces a label-free, prior-free alerting mechanism that automatically detects whether LLM-based adjudicators systematically assign scores below a predefined threshold. Experiments demonstrate its effectiveness in identifying single or multiple anomalous scoring sources, achieving both theoretical rigor—via formal logic and optimization—and practical robustness in unsupervised evaluation.

To address state oscillation and deadlock issues inherent in rule-based decision-making (e.g., finite state machines) for dynamic swarm adversarial scenarios, this paper proposes an interpretable, stable, and adaptive collective intelligence decision framework. Methodologically, it innovatively integrates probabilistic finite state machines (PFSMs), deep convolutional neural networks (CNNs), and multi-agent deep reinforcement learning (MARL): PFSMs ensure decision interpretability and state consistency; CNNs extract high-dimensional environmental features; and MARL enables cooperative policy optimization. Mechanistically, the framework eliminates the root causes of conflicts arising from deterministic state transitions. Experimental results in realistic adversarial settings demonstrate significant improvements in decision stability and task success rate. Specifically, the proposed framework achieves higher human-likeness in collaborative and competitive swarm strategies, superior task completion rates, and enhanced robustness compared to state-of-the-art approaches.

In reinforcement learning, AI-generated preference feedback often suffers from inconsistent judgments—such as preference cycles—leading to unstable training. To address this, we propose the first systematic framework for detecting and eliminating preference conflicts. We introduce the Conflict Detection Rate (CDR) as a quantitative metric for inconsistency and design the Deconflicted Graph Reward (DGR) mechanism: leveraging graph algorithms, DGR transforms the original preference graph into a directed acyclic graph (DAG), thereby generating logically consistent reward signals and purifying rewards prior to policy optimization. Experiments demonstrate that our approach significantly improves training stability and final performance, outperforming strong baselines across multiple benchmarks. This work provides the first empirical validation that logical consistency is a critical determinant of AI feedback quality.

This work addresses the lack of explicit, auditable conflict resolution mechanisms in existing multi-agent requirement negotiation approaches, which hinders compliance with transparency demands in highly regulated contexts. The paper introduces Dung’s abstract argumentation framework into requirements engineering for the first time, modeling proposals, critiques, and refinements as argument nodes, with attack relations capturing conflicts. Acceptable argument sets are derived using grounded and preferred semantics, integrated within a structured negotiation process that combines KAOS goal modeling and multi-layer validation. The approach provides argument-level traceability and enables automatic generation of standards-compliant artifacts. Empirical results demonstrate significantly superior traceability over baselines, higher decision rationality (4.32 vs. 3.07, p<0.001), 94.9% BERTScore semantic retention, and improved compliance coverage at 84.7% compared to baseline ranges of 47.6%–47.8%.

This paper addresses the challenge of achieving reliable consensus among multi-agent systems operating in dynamic environments. Methodologically, it introduces a novel partial-consensus modeling and evaluation framework: (1) formally defining satisfaction and three agreement metrics—minimum, mean, and median agreement degrees; (2) constructing a quantifiable model of value influence; and (3) integrating formal logic with metric analysis within an abstract and value-based argumentation framework to yield a computable agreement assessment tool. Key contributions include: (1) enabling commitment maintenance and decision progression under non-unanimous agreement; (2) providing theoretical proof that partial consensus exhibits robustness under incremental information updates; and (3) integrating the tool into an open-source argumentation reasoning library. Empirical evaluation demonstrates that the framework significantly enhances the naturalness and robustness of multi-agent coordination.

This study addresses the lack of a systematic cognitive framework for managing conflict in uncertain, adversarial, and dynamic environments. It extends the OODA loop—originally developed for aerial combat—into a general-purpose cognitive architecture for cross-domain conflict reasoning, positing that conflict fundamentally entails a contest over observational capacity, judgment, and decision-making capability. By integrating situation awareness, mental models, and goal- and experience-driven reasoning mechanisms, the authors develop an OODA-based cognitive modeling approach. The framework’s efficacy is validated in two distinct scenarios: cyber conflict and cognitive confrontation. Empirical results demonstrate that information processing efficiency and cognitive superiority are pivotal determinants of success in such contested settings.

This study addresses the “correlation-induced consensus blind spot” in multi-agent arbitration, where model convergence can compromise safety monitoring in critical error regions. To mitigate this issue, the authors propose ARAT, a novel system that integrates inductive random forest and analogical k-nearest neighbor agents, augmented with a calibrated meta-model, a conservative coverage strategy, and a safety flag gating mechanism to foster constructive disagreement and alleviate consensus blind spots. Experimental results demonstrate that ARAT reduces the false negative rate from 4.80% to 1.70% on the UNSW-NB15 dataset and validates its efficacy in a cross-domain clinical readmission task. The findings further reveal that enhanced model capability may exacerbate error correlation, underscoring the necessity of diversity oriented toward constructive disagreement to strengthen system safety.

This work addresses structural conflicts between raw data and contextual information by proposing an explicit modeling framework based on generalized operators. The approach formalizes such conflicts as local, directed, and context-dependent mathematical entities, integrating weighting mechanisms, scaling behaviors, and output mappings through a unified abstract operator. In contrast to prior methods that treat conflict merely as a byproduct of optimization, this study is the first to model conflict explicitly as an independent, computable operator at the component level. The resulting framework is agnostic to specific learning algorithms or optimization strategies, offering strong generality and transferability across a wide range of context-sensitive problems.

This work addresses the challenge of conflicting control actions arising from concurrent xApps in Open Radio Access Networks (O-RAN) by proposing a Conflict Resolution Agent (CR Agent) based on deep reinforcement learning. The agent employs an Actor-Critic architecture with the PPO-Clip algorithm to dynamically infer optimal resolution actions by observing network states and conflict-related decision data, leveraging a batch training mechanism in simulation to continually refine its policy. This study represents the first effort to apply deep reinforcement learning to O-RAN conflict management and introduces a systematic evaluation framework for conflict resolution. Experimental results demonstrate that, under medium-to-high traffic conditions, the proposed approach significantly outperforms conventional rule-based baselines, effectively reducing adverse network events caused by control conflicts and thereby validating its efficacy and superiority.

This study addresses the vulnerability of large language models (LLMs) acting as evaluators to post-decision manipulation through interactive challenges, which can compromise judgment reliability. Introducing the novel concept of “post-decision manipulability,” the work proposes an Evaluative Robustness Score (ERS) to quantify LLM stability and directional susceptibility under adversarial interaction. Through controlled experiments, counter-baseline challenge protocols, target-verification balanced designs, and authoritative framework testing—systematically evaluated on MT-Bench and AlpacaEval benchmarks—the findings reveal that while LLM judges remain stable under neutral re-evaluation, their judgments undergo significant reversal when subjected to targeted challenges. This leads to reduced alignment with human preferences, shifts in model rankings, and even high-confidence harmful misjudgments, underscoring critical robustness limitations in current LLM-based evaluation paradigms.