arXiv subjects

Yusheng Hu

Publications and source records attributed to Yusheng Hu.

3 recordsLinked to original sources

arXiv2026

A Divide-and-Conquer Engine for Lexicographical Permutations: Accelerating State Evolution via Hybrid Software-Hardware CPU Instructions

Traditional lexicographical permutation algorithms, epitomized by \texttt{std::next\_permutation}, are fundamentally bottlenecked by dense control logic and frequent branch mispredictions. This paper introduces LexCHA(Lexicographical Co-designed Hardware Acceleration), a novel hardware-software co-designed architecture that accelerates permutation via native SIMD instructions. Exploiting the inherent fractal isomorphism of permutations, LexCHA decouples global state evolution into a macro software unranking phase and a micro hardware block-construction phase. By utilizing pre-computed deterministic mask streams, LexCHA replaces traditional conditional branching with streaming vector shuffles, achieving a branch-free execution flow. Evaluations demonstrate that LexCHA demonstrates significantly higher throughput than scalar implementations of modern processors, outperforming existing state-of-the-art \texttt{std::next\_permutation} implementations significantly.

cs.DS↗

arXiv2025

Dynamic Risk Assessment of Wildland-Urban Interface Fires

Wildland-Urban Interface (WUI) fires represent a compound disaster resulting from the interactions between natural ecosystems and human settlements, characterized by significantly dynamic evolving risks. However, most current risk assessment studies are based on static frameworks, which struggle to effectively capture the dynamic changes in risk over time. To address this issue, this paper proposes an innovative method that integrates a dynamic evaluation matrix, grey incidence analysis, and an optimization model for the dynamic risk assessment of WUI fires. This method incorporates time-series data by constructing a dynamic evaluation matrix, subsequently calculates the weighted standardized matrix for each evaluated area and its local volume matrices relative to the positive and negative ideal matrices. The dynamic differences between the evaluated areas and the ideal state are quantified by calculating the grey incidence degree, and an optimization model is established to solve for the superiority degree used for risk ranking. Research demonstrates that this method not only simplifies the computational process but also effectively captures the dynamic evolution patterns of fire risk across different areas, enabling refined risk classification. Compared to existing static methods, this framework overcomes their limitation in adequately representing risk dynamics, providing a more scientific basis for decision-making in the dynamic management and proactive prevention and control of WUI fires.

q-fin.RM↗

arXiv2025

Pieceformer: Similarity-Driven Knowledge Transfer via Scalable Graph Transformer in VLSI

Accurate graph similarity is critical for knowledge transfer in VLSI design, enabling the reuse of prior solutions to reduce engineering effort and turnaround time. We propose Pieceformer, a scalable, self-supervised similarity assessment framework, equipped with a hybrid message-passing and graph transformer encoder. To address transformer scalability, we incorporate a linear transformer backbone and introduce a partitioned training pipeline for efficient memory and parallelism management. Evaluations on synthetic and real-world CircuitNet datasets show that Pieceformer reduces mean absolute error (MAE) by 24.9% over the baseline and is the only method to correctly cluster all real-world design groups. We further demonstrate the practical usage of our model through a case study on a partitioning task, achieving up to 89% runtime reduction. These results validate the framework's effectiveness for scalable, unbiased design reuse in modern VLSI systems.

cs.LG↗