How Merge-Tolerant Are Vision Transformers for Wheat Phenotyping?
研究通过对比不同视觉变换器在小麦表型任务中的性能，探讨了无训练令牌合并方法对提高处理速度和减少资源消耗的效果。
0 citationsRead paper
Selected work
Representative Papers
Top-$k$ Pareto Bandits: Hypervolume Regret for Multi-Objective Slate Selection
This work proposes a novel representation learning framework that addresses the limited representational capacity of existing methods in complex scenarios by integrating adaptive multi-scale fusion with contrastive learning. The approach dynamically aggregates multi-level semantic information and introduces a structure-aware contrastive loss, thereby significantly enhancing the model’s ability to discriminate fine-grained differences. Extensive experiments demonstrate that the proposed framework consistently outperforms state-of-the-art methods across multiple benchmark datasets, exhibiting particularly strong robustness under low-resource settings and in the presence of noise. Beyond advancing the theoretical foundations of representation learning, this study also delivers an efficient and scalable solution with practical applicability.
0 citationsRead paper
Graph-Series Semantics and Abel Regularization for Recursive Hybrid Quantum Programs
This work addresses the challenge of modeling infinite execution and convergence in recursive hybrid quantum programs by proposing a semantic framework based on graded graph sequences. Finite terminating executions are characterized via directed paths, while infinite recursive unfoldings are handled through Abel regularization. Quantum-classical interactions are uniformly described using the quantum orchestra monad. The main contributions include the first integration of graph-sequence semantics with Abel regularization, the establishment of an exact correspondence between execution graphs and both Kleene approximations and least fixed-point semantics, and the introduction of the Fredholm feedback determinant to detect singularities in recursive structures. Theoretical results confirm that the proposed semantics aligns with the standard fixed-point model and that the regularized semantics converges in the limit, thereby effectively supporting the identification of singular recursive configurations.
0 citationsRead paper
Position Spaces and Graphs
This study addresses the formal modeling of relative positional relationships among discrete symbols, balancing logical consistency with computational tractability. To this end, the authors propose a “position graph” framework grounded in positional space theory, employing two strict partial orders to capture horizontal and vertical alignment and precedence relations. Symbol arrangements are constrained by chain conditions and row-column compatibility requirements. The model preserves expressive power while guaranteeing rigorous algebraic consistency. Key contributions include establishing necessary and sufficient conditions for position graph consistency, proving that the induced subgraph isomorphism problem is NP-complete—thereby revealing the intrinsic computational complexity of structural pattern discovery—and constructing a formal logical layer independent of specific data extraction techniques.
0 citationsRead paper
Stein Variational Black-Box Combinatorial Optimization
This work addresses the challenge of premature convergence to a single optimum in high-dimensional combinatorial black-box optimization, where balancing exploration and exploitation remains difficult. The authors propose the first integration of Stein variational gradient descent into estimation-of-distribution algorithms, introducing a repulsive mechanism among particles in the parameter space to drive the population toward collaborative exploration of multiple fitness peaks. This approach significantly enhances multimodal search capability, achieving performance that matches or surpasses state-of-the-art algorithms across a range of benchmark problems, with particularly notable improvements on large-scale instances.
0 citationsRead paper
Recent publications
Latest Papers
How Merge-Tolerant Are Vision Transformers for Wheat Phenotyping?
研究通过对比不同视觉变换器在小麦表型任务中的性能，探讨了无训练令牌合并方法对提高处理速度和减少资源消耗的效果。
0 citationsRead paper
Top-$k$ Pareto Bandits: Hypervolume Regret for Multi-Objective Slate Selection
This work proposes a novel representation learning framework that addresses the limited representational capacity of existing methods in complex scenarios by integrating adaptive multi-scale fusion with contrastive learning. The approach dynamically aggregates multi-level semantic information and introduces a structure-aware contrastive loss, thereby significantly enhancing the model’s ability to discriminate fine-grained differences. Extensive experiments demonstrate that the proposed framework consistently outperforms state-of-the-art methods across multiple benchmark datasets, exhibiting particularly strong robustness under low-resource settings and in the presence of noise. Beyond advancing the theoretical foundations of representation learning, this study also delivers an efficient and scalable solution with practical applicability.
0 citationsRead paper
Graph-Series Semantics and Abel Regularization for Recursive Hybrid Quantum Programs
This work addresses the challenge of modeling infinite execution and convergence in recursive hybrid quantum programs by proposing a semantic framework based on graded graph sequences. Finite terminating executions are characterized via directed paths, while infinite recursive unfoldings are handled through Abel regularization. Quantum-classical interactions are uniformly described using the quantum orchestra monad. The main contributions include the first integration of graph-sequence semantics with Abel regularization, the establishment of an exact correspondence between execution graphs and both Kleene approximations and least fixed-point semantics, and the introduction of the Fredholm feedback determinant to detect singularities in recursive structures. Theoretical results confirm that the proposed semantics aligns with the standard fixed-point model and that the regularized semantics converges in the limit, thereby effectively supporting the identification of singular recursive configurations.
0 citationsRead paper
Position Spaces and Graphs
This study addresses the formal modeling of relative positional relationships among discrete symbols, balancing logical consistency with computational tractability. To this end, the authors propose a “position graph” framework grounded in positional space theory, employing two strict partial orders to capture horizontal and vertical alignment and precedence relations. Symbol arrangements are constrained by chain conditions and row-column compatibility requirements. The model preserves expressive power while guaranteeing rigorous algebraic consistency. Key contributions include establishing necessary and sufficient conditions for position graph consistency, proving that the induced subgraph isomorphism problem is NP-complete—thereby revealing the intrinsic computational complexity of structural pattern discovery—and constructing a formal logical layer independent of specific data extraction techniques.
0 citationsRead paper
Stein Variational Black-Box Combinatorial Optimization
This work addresses the challenge of premature convergence to a single optimum in high-dimensional combinatorial black-box optimization, where balancing exploration and exploitation remains difficult. The authors propose the first integration of Stein variational gradient descent into estimation-of-distribution algorithms, introducing a repulsive mechanism among particles in the parameter space to drive the population toward collaborative exploration of multiple fitness peaks. This approach significantly enhances multimodal search capability, achieving performance that matches or surpasses state-of-the-art algorithms across a range of benchmark problems, with particularly notable improvements on large-scale instances.
0 citationsRead paper