LEAP: Learning Emergent Active Perception for Quadruped Navigation
本文提出LEAP方法，通过目标导向导航和自我中心信念图来学习主动感知，以解决四足机器人在危险地形中导航的问题。
0 citationsRead paper
Selected work
Representative Papers
Streaming algorithms for computing coresets and $k$-median clustering in the Hamming space
本文提出了一种在Hamming空间中计算连续k-中位数聚类问题的(1+ε)近似算法，该算法具有FPT运行时间，并可在流式处理中实现。
0 citationsRead paper
Structured Prediction for Scalable Spreadsheet Table Understanding: From Cell Types to Table Ranges (Extended Version)
This study addresses the challenge of structured understanding in spreadsheets caused by heterogeneous layouts. We propose a two-stage pipeline integrating CRF-LightGBM with deterministic algorithms and introduce the StatSheets benchmark dataset. By leveraging a lightweight model alongside a five-stage range extraction algorithm, our approach achieves predictive performance comparable to Transformer-based models at significantly lower computational costs. Experimental results demonstrate a cell classification F1-score of 0.937 and superior table detection accuracy over existing baselines, with markedly higher computational efficiency. Consequently, this work establishes an efficient and scalable paradigm for large-scale spreadsheet parsing, effectively balancing high performance with resource constraints in document layout analysis.
0 citationsRead paper
Limit Points of Reflow with Minibatch Optimal Transport
This work investigates the asymptotic behavior of alternating Reflow iterations with mini-batch optimal transport (OT) under a fixed batch size. By introducing a novel notion of weakly corrected couplings, the authors prove that the limiting coupling satisfies N-cyclical monotonicity and, under suitable support conditions, coincides with the optimal transport map between the endpoint distributions. The analysis integrates the Reflow framework, mini-batch OT, N-cyclical monotonicity, and gradient-field constraints to demonstrate that the limit exhibits favorable structural properties—such as correctability and linearity. This study establishes, for the first time, a rigorous theoretical connection between the Reflow limit and classical optimal transport theory, thereby providing a formal convergence guarantee for the Reflow process.
0 citationsRead paper
Bidirectional Typing with Freezing, Skeletons, and Ghosts
Traditional bidirectional type inference struggles to support first-class polymorphism and faces limitations in balancing flexibility and predictability in mixed information flow. This work proposes Fresco, the first approach integrating skeleton, phantom, and freeze mechanisms: skeletons enable localized bidirectional type information flow between functions and arguments, phantoms represent unknown types, and the freeze mechanism allows users to explicitly control the direction of information flow. Fresco supports flexible yet predictable type inference for first-class polymorphism and extends naturally to modal effect type systems. We present a concise declarative specification along with a corresponding algorithm, prove its soundness and completeness relative to the declarative system, and demonstrate through a prototype implementation that Fresco offers superior expressiveness, controllability, and extensibility.
0 citationsRead paper
Recent publications
Latest Papers
LEAP: Learning Emergent Active Perception for Quadruped Navigation
本文提出LEAP方法，通过目标导向导航和自我中心信念图来学习主动感知，以解决四足机器人在危险地形中导航的问题。
0 citationsRead paper
Streaming algorithms for computing coresets and $k$-median clustering in the Hamming space
本文提出了一种在Hamming空间中计算连续k-中位数聚类问题的(1+ε)近似算法，该算法具有FPT运行时间，并可在流式处理中实现。
0 citationsRead paper
Structured Prediction for Scalable Spreadsheet Table Understanding: From Cell Types to Table Ranges (Extended Version)
This study addresses the challenge of structured understanding in spreadsheets caused by heterogeneous layouts. We propose a two-stage pipeline integrating CRF-LightGBM with deterministic algorithms and introduce the StatSheets benchmark dataset. By leveraging a lightweight model alongside a five-stage range extraction algorithm, our approach achieves predictive performance comparable to Transformer-based models at significantly lower computational costs. Experimental results demonstrate a cell classification F1-score of 0.937 and superior table detection accuracy over existing baselines, with markedly higher computational efficiency. Consequently, this work establishes an efficient and scalable paradigm for large-scale spreadsheet parsing, effectively balancing high performance with resource constraints in document layout analysis.
0 citationsRead paper
Limit Points of Reflow with Minibatch Optimal Transport
This work investigates the asymptotic behavior of alternating Reflow iterations with mini-batch optimal transport (OT) under a fixed batch size. By introducing a novel notion of weakly corrected couplings, the authors prove that the limiting coupling satisfies N-cyclical monotonicity and, under suitable support conditions, coincides with the optimal transport map between the endpoint distributions. The analysis integrates the Reflow framework, mini-batch OT, N-cyclical monotonicity, and gradient-field constraints to demonstrate that the limit exhibits favorable structural properties—such as correctability and linearity. This study establishes, for the first time, a rigorous theoretical connection between the Reflow limit and classical optimal transport theory, thereby providing a formal convergence guarantee for the Reflow process.
0 citationsRead paper
Bidirectional Typing with Freezing, Skeletons, and Ghosts
Traditional bidirectional type inference struggles to support first-class polymorphism and faces limitations in balancing flexibility and predictability in mixed information flow. This work proposes Fresco, the first approach integrating skeleton, phantom, and freeze mechanisms: skeletons enable localized bidirectional type information flow between functions and arguments, phantoms represent unknown types, and the freeze mechanism allows users to explicitly control the direction of information flow. Fresco supports flexible yet predictable type inference for first-class polymorphism and extends naturally to modal effect type systems. We present a concise declarative specification along with a corresponding algorithm, prove its soundness and completeness relative to the declarative system, and demonstrate through a prototype implementation that Fresco offers superior expressiveness, controllability, and extensibility.
0 citationsRead paper