arXiv subjects

Mengying Zhao

Publications and source records attributed to Mengying Zhao.

5 recordsLinked to original sources

arXiv2026

Porting Declarative UI to HarmonyOS: A Heuristic-guided LLM Approach

As an emerging operating system, HarmonyOS has a significant demand for software migration from platforms such as Android and iOS, where the User Interface (UI) translation accounts for a critical link. However, the latest UI development has shifted to declarative paradigms, e.g., Kotlin Jetpack Compose (KJC) for Android, SwiftUI for iOS, and ArkUI for HarmonyOS, rendering prior translation approaches inapplicable, as they target either backend logic or legacy imperative UIs. As such, this paper targets ArkUI and proposes an automatic translation approach, namely ArkTrans, to port UI files from Android and iOS to HarmonyOS. ArkTrans overcomes two salient challenges during the translation: (1) Programming Language (PL) unfamiliarity, and (2) severe syntactic chaos. Towards the first challenge, ArkTrans heuristically constructs ArkUI skeletons by extracting metadata from source PL, thereby guiding LLMs' initial translation. As for the second challenge, ArkTrans executes empirically revealed post-fixing rules via pattern matching to repair most of the remaining syntactic errors. To examine the effectiveness of ArkTrans, we construct a 100-sample parallel UI page translation benchmark from KJC/SwiftUI to ArkUI at the file level. Extensive experiments demonstrate that LLMs with direct/one-shot prompting cannot translate a single compilable UI page. In contrast, at most 90.67\% ArkTrans-translated files can be successfully compiled with high visual fidelity.

cs.SE↗

arXiv2026

Semantic-Geometric Dual Compression: Training-Free Visual Token Reduction for Ultra-High-Resolution Remote Sensing Understanding

Multimodal Large Language Models (MLLMs) have demonstrated immense potential in Earth observation. However, the massive visual tokens generated when processing Ultra-High-Resolution (UHR) imagery introduce prohibitive computational overhead, severely bottlenecking their inference efficiency. Existing visual token compression methods predominantly adopt static and uniform compression strategies, neglecting the inherent "Semantic-Geometric Duality" in remote sensing interpretation tasks. Specifically, object semantic tasks focus on the abstract semantics of objects and benefit from aggressive background pruning, whereas scene geometric tasks critically rely on the integrity of spatial topology. To address this challenge, we propose DualComp, a task-adaptive dual-stream token compression framework. Dynamically guided by a lightweight pre-trained router, DualComp decouples feature processing into two dedicated pathways. In the object semantic stream, the Spatially-Contiguous Semantic Aggregator (SCSA) utilizes size-adaptive clustering to aggregates redundant background while protecting small object. In the scene geometric stream, the Instruction-Guided Structure Recoverer (IGSR) introduces a greedy path-tracing topology completion mechanism to reconstruct spatial skeletons. Experiments on the UHR remote sensing benchmark XLRS-Bench demonstrate that DualComp accomplishes high-fidelity remote sensing interpretation at an exceptionally low computational cost, achieving simultaneous improvements in both efficiency and accuracy.

cs.CV↗

arXiv2026

Extended Gabidulin-Kronecker Product Codes and Their Application to Cryptosystems

In this paper, we initiate the study of Extended Gabidulin codes with a Kronecker product structure and propose three enhanced variants of the Rank Quasi-Cyclic (RQC) (Melchor et.al., IEEE IT, 2018) cryptosystem. First, we establish precise bounds on the minimum rank distance of Gabidulin-Kronecker product codes under two distinct parameter regimes. Specifically, when $n_{1}=k_{1}$ and $n_{2}=m<n_{1}n_{2}$, the minimum rank distance is exactly $n_{2}-k_{2}+1$. This yields a new family of Maximum Rank Distance (MRD) codes, which are distinct from classical Gabidulin codes. For the case of $k_{1}\leq n_{1},k_{2}\leq n_{2},n_{1}n_{2}\leq m$, the minimum rank distance $d$ of Gabidulin-Kronecker product codes satisfies a tight upper and lower bound, i.e., $n_{2}-k_{2}+1 \leq d \leq (n_{1}-k_{1}+1)(n_{2}-k_{2}+1)$. Second, we introduce a new class of decodable rank-metric codes, namely Extended Gabidulin-Kronecker product (EGK) codes, which generalize the structure of Gabidulin-Kronecker product (GK) codes. We also propose a decoding algorithm that directly retrieves the codeword without recovering the error vector, thus improving efficiency. This algorithm achieves zero decoding failure probability when the error weight is within its correction capability. Third, we propose three enhanced variants of the RQC cryptosystem based on EGK codes, each offering a distinct trade-off between security and efficiency. For 128-bit security, all variants achieve significant reductions in public key size compared to the Multi-UR-AG (Bidoux et.al., IEEE IT, 2024) while ensuring zero decryption failure probability--a key security advantage over many existing rank-based schemes.

cs.IT↗

arXiv2025

SysVCoder: An LLM-Driven Framework for Systematic Generation of System-Level Design

Recent advances in large language models (LLMs) have demonstrated strong potential in generating hardware designs using hardware description languages (HDLs) such as Verilog. However, existing LLM-based frameworks struggle to accurately capture the complexity of real-world architectural designs, particularly for large-scale systems with hierarchical, multi-level module instantiations. To address this issue, we present SysVCoder, an LLM-driven framework that enhances both the generation quality and efficiency of system-level design in Verilog. SysVCoder introduces a two-stage generation pipeline that leverages an intermediate representation to enable a more structured and accurate translation from natural language specifications to complex multi-module designs. Furthermore, we incorporate a rule-based alignment mechanism and a domain-specific retrieval-augmented generation strategy (DS-RAG) to enhance functional correctness by grounding LLM outputs in domain knowledge. We also present SysVDB, a comprehensive dataset comprising 60 system-level hardware designs along with their corresponding verification testbenches. Experimental results demonstrate that SysVCoder outperforms state-of-the-art frameworks such as CodeV and VeriGen by 30.7% and 38.3% in terms of functional correctness under the same base LLM. Notably, SysVCoder achieves performance comparable to NVIDIA's GPT-4 based VerilogCoder while using only a 7B-parameter model, reducing token consumption by 7.6x and synthesis latency by 37.5x. Both SysVCoder and SysVDB are made public at https://gitee.com/sdu-aes-lab/sysvcoder/.

cs.SE↗

arXiv2024

Unity is Power: Semi-Asynchronous Collaborative Training of Large-Scale Models with Structured Pruning in Resource-Limited Clients

In this work, we study to release the potential of massive heterogeneous weak computing power to collaboratively train large-scale models on dispersed datasets. In order to improve both efficiency and accuracy in resource-adaptive collaborative learning, we take the first step to consider the \textit{unstructured pruning}, \textit{varying submodel architectures}, \textit{knowledge loss}, and \textit{straggler} challenges simultaneously. We propose a novel semi-asynchronous collaborative training framework, namely ${Co\text{-}S}^2{P}$, with data distribution-aware structured pruning and cross-block knowledge transfer mechanism to address the above concerns. Furthermore, we provide theoretical proof that ${Co\text{-}S}^2{P}$ can achieve asymptotic optimal convergence rate of $O(1/\sqrt{N^*EQ})$. Finally, we conduct extensive experiments on two types of tasks with a real-world hardware testbed including diverse IoT devices.The experimental results demonstrate that $Co\text{-}S^2P$ improves accuracy by up to 8.8\% and resource utilization by up to 1.2$\times$ compared to state-of-the-art methods, while reducing memory consumption by approximately 22\% and training time by about 24\% on all resource-limited devices.

cs.DC↗