SV-WAM: An Efficient Surround-View World-Action Model for End-to-End Autonomous Driving
为解决自动驾驶中全视角覆盖与计算效率的问题，SV-WAM通过共享生成模型进行动作学习，并采用可行驶区域合规正则化器提高安全性。
0 citationsRead paper
Selected work
Representative Papers
Can Large Language Models Explain Flight Safety Events? A Prior-Guided Semantic LLM-based Approach
研究提出FlightLLM，一种基于先验引导的语义大语言模型方法，通过特征工程、语义离散化及少量样本对比学习等手段解决飞行安全事件解释问题。
0 citationsRead paper
Cross-view geo-localization, Image retrieval, Multiscale geometric modeling, Frequency domain enhancement
This work addresses the challenges of cross-view geolocalization, including large viewpoint discrepancies, inconsistent textures, and degraded local discriminative cues. To this end, the authors propose the Spatial and Frequency Domain Enhancement network (SFDE), which introduces frequency-domain invariance into this task for the first time. SFDE employs a parallel three-branch architecture to jointly model global semantics, local geometric structures, and frequency-domain statistical properties, enabling multi-granularity consistent representations. By integrating multi-scale geometric modeling with progressive enhancement and coupled constraint optimization within a unified embedding space, the method achieves state-of-the-art or competitive performance across multiple benchmarks, significantly improving retrieval and localization accuracy while maintaining a lightweight and efficient design.
0 citationsRead paper
Multilevel constructions of constant dimension codes based on one-factorization of complete graphs
This paper addresses the low lower bound on minimum distance in constant-dimension codes (CDCs). To tackle this, we propose a multi-level construction method based on one-factorizations of complete graphs. Our key innovation lies in designing novel skeleton codes via binary vector transformations induced by one-factorizations, and integrating quasi-pendant block structures with optimal Ferrers diagram rank-metric codes to construct CDCs achieving high minimum distance. The approach systematically refines the multi-level construction framework: while fixing the dimension $k = 6$ and minimum distance $d = 8$, it significantly improves the cardinality lower bounds for code lengths $n in [16,19]$, thereby advancing the state-of-the-art lower bounds on $overline{A}_q(n,8,6)$. This work provides both new theoretical insights and practical construction tools for network coding and random linear network coding over finite fields.
0 citationsRead paper
New constructions of cyclic constant-dimension subspace codes based on Sidon spaces and subspace polynomials
This work addresses the size limitation of cyclic constant-dimension codes (cyclic CDCs). We propose a novel construction framework based on flexible parametrized Sidon spaces and subspace polynomials. By designing two scalable families of Sidon spaces and leveraging finite-field algebraic structures, our method significantly improves codebook size and parameter adaptability—particularly achieving breakthroughs for critical ambient space dimensions $n = (2r+1)k$ and $n = 2rk$. The resulting cyclic CDCs attain minimum subspace distance $2k-2$, and for $n = 4k$, their asymptotic code rate reaches half the sphere-packing bound—surpassing all known optimal constructions in cardinality. Crucially, this is the first systematic integration of Sidon spaces into cyclic CDC design, establishing a new paradigm for high-dimensional network coding that bridges theoretical rigor and practical applicability.
0 citationsRead paper
Recent publications
Latest Papers
SV-WAM: An Efficient Surround-View World-Action Model for End-to-End Autonomous Driving
为解决自动驾驶中全视角覆盖与计算效率的问题，SV-WAM通过共享生成模型进行动作学习，并采用可行驶区域合规正则化器提高安全性。
0 citationsRead paper
Can Large Language Models Explain Flight Safety Events? A Prior-Guided Semantic LLM-based Approach
研究提出FlightLLM，一种基于先验引导的语义大语言模型方法，通过特征工程、语义离散化及少量样本对比学习等手段解决飞行安全事件解释问题。
0 citationsRead paper
Cross-view geo-localization, Image retrieval, Multiscale geometric modeling, Frequency domain enhancement
This work addresses the challenges of cross-view geolocalization, including large viewpoint discrepancies, inconsistent textures, and degraded local discriminative cues. To this end, the authors propose the Spatial and Frequency Domain Enhancement network (SFDE), which introduces frequency-domain invariance into this task for the first time. SFDE employs a parallel three-branch architecture to jointly model global semantics, local geometric structures, and frequency-domain statistical properties, enabling multi-granularity consistent representations. By integrating multi-scale geometric modeling with progressive enhancement and coupled constraint optimization within a unified embedding space, the method achieves state-of-the-art or competitive performance across multiple benchmarks, significantly improving retrieval and localization accuracy while maintaining a lightweight and efficient design.
0 citationsRead paper
Multilevel constructions of constant dimension codes based on one-factorization of complete graphs
This paper addresses the low lower bound on minimum distance in constant-dimension codes (CDCs). To tackle this, we propose a multi-level construction method based on one-factorizations of complete graphs. Our key innovation lies in designing novel skeleton codes via binary vector transformations induced by one-factorizations, and integrating quasi-pendant block structures with optimal Ferrers diagram rank-metric codes to construct CDCs achieving high minimum distance. The approach systematically refines the multi-level construction framework: while fixing the dimension $k = 6$ and minimum distance $d = 8$, it significantly improves the cardinality lower bounds for code lengths $n in [16,19]$, thereby advancing the state-of-the-art lower bounds on $overline{A}_q(n,8,6)$. This work provides both new theoretical insights and practical construction tools for network coding and random linear network coding over finite fields.
0 citationsRead paper
New constructions of cyclic constant-dimension subspace codes based on Sidon spaces and subspace polynomials
This work addresses the size limitation of cyclic constant-dimension codes (cyclic CDCs). We propose a novel construction framework based on flexible parametrized Sidon spaces and subspace polynomials. By designing two scalable families of Sidon spaces and leveraging finite-field algebraic structures, our method significantly improves codebook size and parameter adaptability—particularly achieving breakthroughs for critical ambient space dimensions $n = (2r+1)k$ and $n = 2rk$. The resulting cyclic CDCs attain minimum subspace distance $2k-2$, and for $n = 4k$, their asymptotic code rate reaches half the sphere-packing bound—surpassing all known optimal constructions in cardinality. Crucially, this is the first systematic integration of Sidon spaces into cyclic CDC design, establishing a new paradigm for high-dimensional network coding that bridges theoretical rigor and practical applicability.
0 citationsRead paper