Do Large Language Models Possess a Theory of Mind? A Comparative Evaluation Using the Strange Stories Paradigm
研究通过改编的故事测试法比较了五种大型语言模型与人类在心智理论能力上的表现,以探讨这些模型是否能从文本中推断出他人的信念、意图和情感。
研究通过改编的故事测试法比较了五种大型语言模型与人类在心智理论能力上的表现,以探讨这些模型是否能从文本中推断出他人的信念、意图和情感。
本文提出MiNCE框架,利用RKHS理论构建带限函数及其平滑谱的一致性置信包络,并证明了其在无噪声和有噪声观测模型下的强一致性。
本文通过量子计算范式及经典张量网络等效方法,特别是使用基于矩阵乘积态的求解器,解决了二维声波方程的大规模模拟问题。
研究解决了四项二元多项式乘法的布尔乘法复杂性问题,通过结构化证明方法确定其复杂性为九,并验证了结果。
This study addresses weighted fair division and discrepancy theory under matroid constraints, proposing strongly polynomial-time algorithms grounded in local exchange theorems and constructive proofs. Key contributions include establishing weighted matroid partitionability and extending the Beck-Fiala framework to matroid settings, yielding logarithmic discrepancy bounds. The work achieves EF1 and additive approximation guarantees for fair allocation and scheduling optimization while refuting the weighted carpool conjecture. By systematically resolving fairness and discrepancy control challenges in constrained environments, this research provides novel theoretical foundations and efficient algorithmic tools for related combinatorial optimization problems.
本文提出MiNCE框架,利用RKHS理论构建带限函数及其平滑谱的一致性置信包络,并证明了其在无噪声和有噪声观测模型下的强一致性。
本文通过量子计算范式及经典张量网络等效方法,特别是使用基于矩阵乘积态的求解器,解决了二维声波方程的大规模模拟问题。
研究解决了四项二元多项式乘法的布尔乘法复杂性问题,通过结构化证明方法确定其复杂性为九,并验证了结果。
This study addresses weighted fair division and discrepancy theory under matroid constraints, proposing strongly polynomial-time algorithms grounded in local exchange theorems and constructive proofs. Key contributions include establishing weighted matroid partitionability and extending the Beck-Fiala framework to matroid settings, yielding logarithmic discrepancy bounds. The work achieves EF1 and additive approximation guarantees for fair allocation and scheduling optimization while refuting the weighted carpool conjecture. By systematically resolving fairness and discrepancy control challenges in constrained environments, this research provides novel theoretical foundations and efficient algorithmic tools for related combinatorial optimization problems.
This work addresses the limitations of conventional human-robot collaboration systems, which are constrained by tethered connections, lack modularity and rapid reconfigurability, and fail to meet real-time perception and safety requirements with existing commercial wireless solutions. The authors propose an infrastructure-free, 5G-enabled wireless reconfigurable collaborative unit architecture integrating a battery-powered multi-sensor platform and an edge vision module, enabling cross-scenario deployment. By training a highly robust hand and grasp pose estimation model using a fusion of synthetic and real-world data, and leveraging 5G edge computing to optimize the bandwidth–latency trade-off, the system achieves a round-trip latency of 12 ms across multiple international 5G networks, a mean average precision (mAP@50–95) of 97.74% ± 0.10% for pose detection, and an average inference time of 12.5 ms, thereby demonstrating the feasibility of safe and adaptive human-robot collaboration.