Surprise Reduction and Nullification in Bayesian and Inverse Bayesian Inference under Ambiguous Prediction-Error Attribution
研究解决了非平稳环境中预测误差归因问题，通过贝叶斯和逆贝叶斯推理方法区分并形式化了惊讶减少与消除策略。
0 citationsRead paper
Selected work
Representative Papers
Approximation Algorithms for Perfect Fair-Triangle Packing
This work introduces and investigates the fair maximum-weight triangle packing problem: given 3n vertices colored red and blue, the goal is to partition them into n vertex-disjoint triangles such that each triangle contains at least one red and one blue vertex, while maximizing the total edge weight. The problem is shown to be NP-hard. The authors propose two polynomial-time approximation algorithms: a deterministic 1/3-approximation algorithm based on matching and maximum-weight [1,2]-factor computation running in O(n³) time, and an improved randomized (16/47 − ε)-approximation algorithm combining random cycle-breaking with maximum-weight matching, which runs in O(n⁴) time. These results substantially advance the tractability frontier for this class of fair combinatorial optimization problems.
0 citationsRead paper
Cybernetic Android Avatar "Yui": System Integration, Field Deployment, and Evaluation
This work addresses the challenge of establishing a shared sense of physical co-presence and effectively conveying nonverbal social cues in remote interaction. To this end, the authors present “Yui,” a highly realistic full-body avatar integrating a 55-degree-of-freedom anthropomorphic mechanism, expressive facial and gaze control, dexterous upper-limb manipulation, and a mobile base, supporting both head-mounted and desktop teleoperation modes. The system was deployed for the first time in diverse real-world settings—including long-term public exhibitions and educational exchanges—with a cumulative operational duration of 1,131 hours. Findings indicate that users consistently perceived strong co-presence, human-likeness, and clarity in emotional intent communication, confirming the system’s usability for general audiences, while also highlighting the need for improved control precision. These insights offer critical design guidance for the practical deployment of embodied social robots.
0 citationsRead paper
Computational Implementation of a Model of Category-Theoretic Metaphor Comprehension
This study addresses the longstanding challenge in metaphor comprehension of simultaneously accounting for both systematicity and novelty. Building on Fuyama et al.’s Theory of Indeterminate Natural Transformations (TINT), the work presents the first computationally implementable model that adheres more closely to the original categorical formulation. By streamlining TINT’s computational architecture, the authors develop an efficient algorithm and validate it through fitting and simulation against human experimental data. The results demonstrate that the proposed model significantly outperforms existing approaches across three critical dimensions: goodness-of-fit to empirical data, systematicity, and capacity for handling novel metaphors. This advancement offers a new theoretical and practical pathway for computational cognitive modeling of figurative language understanding.
0 citationsRead paper
MoireMix: A Formula-Based Data Augmentation for Improving Image Classification Robustness
This work proposes a novel analytic data augmentation method for image classification based on moiré interference patterns, addressing the high computational cost and reliance on external data common in existing approaches such as diffusion models or complex feature mixing. By introducing closed-form, multi-scale structured perturbations directly into the training pipeline, the method requires no external data, incurs negligible storage overhead, and achieves extremely low computational cost—synthesizing and discarding each perturbation on-the-fly during training. Integrated with Vision Transformers, the approach significantly outperforms current data-free augmentation techniques on robustness benchmarks including ImageNet-C, ImageNet-R, and adversarial evaluations, generating perturbations for a single image in just 0.0026 seconds.
0 citationsRead paper
Recent publications
Latest Papers
Surprise Reduction and Nullification in Bayesian and Inverse Bayesian Inference under Ambiguous Prediction-Error Attribution
研究解决了非平稳环境中预测误差归因问题，通过贝叶斯和逆贝叶斯推理方法区分并形式化了惊讶减少与消除策略。
0 citationsRead paper
Approximation Algorithms for Perfect Fair-Triangle Packing
This work introduces and investigates the fair maximum-weight triangle packing problem: given 3n vertices colored red and blue, the goal is to partition them into n vertex-disjoint triangles such that each triangle contains at least one red and one blue vertex, while maximizing the total edge weight. The problem is shown to be NP-hard. The authors propose two polynomial-time approximation algorithms: a deterministic 1/3-approximation algorithm based on matching and maximum-weight [1,2]-factor computation running in O(n³) time, and an improved randomized (16/47 − ε)-approximation algorithm combining random cycle-breaking with maximum-weight matching, which runs in O(n⁴) time. These results substantially advance the tractability frontier for this class of fair combinatorial optimization problems.
0 citationsRead paper
Cybernetic Android Avatar "Yui": System Integration, Field Deployment, and Evaluation
This work addresses the challenge of establishing a shared sense of physical co-presence and effectively conveying nonverbal social cues in remote interaction. To this end, the authors present “Yui,” a highly realistic full-body avatar integrating a 55-degree-of-freedom anthropomorphic mechanism, expressive facial and gaze control, dexterous upper-limb manipulation, and a mobile base, supporting both head-mounted and desktop teleoperation modes. The system was deployed for the first time in diverse real-world settings—including long-term public exhibitions and educational exchanges—with a cumulative operational duration of 1,131 hours. Findings indicate that users consistently perceived strong co-presence, human-likeness, and clarity in emotional intent communication, confirming the system’s usability for general audiences, while also highlighting the need for improved control precision. These insights offer critical design guidance for the practical deployment of embodied social robots.
0 citationsRead paper
Computational Implementation of a Model of Category-Theoretic Metaphor Comprehension
This study addresses the longstanding challenge in metaphor comprehension of simultaneously accounting for both systematicity and novelty. Building on Fuyama et al.’s Theory of Indeterminate Natural Transformations (TINT), the work presents the first computationally implementable model that adheres more closely to the original categorical formulation. By streamlining TINT’s computational architecture, the authors develop an efficient algorithm and validate it through fitting and simulation against human experimental data. The results demonstrate that the proposed model significantly outperforms existing approaches across three critical dimensions: goodness-of-fit to empirical data, systematicity, and capacity for handling novel metaphors. This advancement offers a new theoretical and practical pathway for computational cognitive modeling of figurative language understanding.
0 citationsRead paper
MoireMix: A Formula-Based Data Augmentation for Improving Image Classification Robustness
This work proposes a novel analytic data augmentation method for image classification based on moiré interference patterns, addressing the high computational cost and reliance on external data common in existing approaches such as diffusion models or complex feature mixing. By introducing closed-form, multi-scale structured perturbations directly into the training pipeline, the method requires no external data, incurs negligible storage overhead, and achieves extremely low computational cost—synthesizing and discarding each perturbation on-the-fly during training. Integrated with Vision Transformers, the approach significantly outperforms current data-free augmentation techniques on robustness benchmarks including ImageNet-C, ImageNet-R, and adversarial evaluations, generating perturbations for a single image in just 0.0026 seconds.
0 citationsRead paper