arXiv subjects

Xinzheng Lu

Publications and source records attributed to Xinzheng Lu.

6 recordsLinked to original sources

arXiv2026

StructureClaw: Traceable LLM Agents and an Executable Benchmark for Structural Engineering Workflows

Addressing a structural-engineering request requires more than a single answer; it requires a chain of interdependent artifacts: interpreted requirements, a computable model, validation records, solver outputs, applicable engineering checks, and a final report. Evaluations centered on question answering or script generation may therefore reward fluent outputs even when the underlying workflow is incomplete, inconsistent, or non-executable. We present StructureClaw, an artifact-centered workbench in which LLM agents operate through governed engineering skills, typed tools, shared artifact state, and local analysis backends, together with StructureClaw-Bench, an executable benchmark of 150 controlled scenarios spanning standard workflows, interactive robustness, and multimodal structural-model reconstruction. Its analyzable standard and multimodal cases require both strict one-to-one structural-model matching and numerical-response agreement with frozen reference responses from the selected analysis engine; interactive cases instead require positive clarification or recovery evidence together with safe non-execution when appropriate. A trial succeeds only when every fixture-required assertion passes. Across nine text-agent configurations, generic-only execution passed the model-artifact check in 87.0% of retained outcomes but achieved only 22.0% E2E Success, whereas automatic StructureClaw reached 82.9%. Interactive and multimodal evaluations further identify semantic state consistency and executable model reconstruction as the dominant remaining bottlenecks. The code and benchmark are available at https://github.com/structureclaw/structureclaw.

cs.SE↗

arXiv2026

Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans

Architectural floor plan design demands joint reasoning over geometry, semantics, and spatial hierarchy, which remains a major challenge for current AI systems. Although recent diffusion and language models improve visual fidelity, they still struggle with coherent spatial reasoning and controllable generation. We present HouseMind, a multimodal large language model that unifies floor plan understanding, generation, and editing in one framework. We introduce discrete room-instance tokens to construct a unified vocabulary that bridges layouts and symbolic reasoning. With multimodal alignment and instruction tuning, the model synthesizes coherent, controllable layouts from text instructions. Experiments show how the framework achieves superior geometric validity and controllability while remaining efficient and locally deployable.

cs.CV↗

arXiv2024

QuakeFormer: A Uniform Approach to Earthquake Ground Motion Prediction Using Masked Transformers

Ground motion prediction (GMP) models are critical for hazard reduction before, during and after destructive earthquakes. In these three stages, intensity forecasting, early warning and interpolation models are corresponding employed to assess the risk. Considering the high cost in numerical methods and the oversimplification in statistical methods, deep-learning-based approaches aim to provide accurate and near-real-time ground motion prediction. Current approaches are limited by specialized architectures, overlooking the interconnection among these three tasks. What's more, the inadequate modeling of absolute and relative spatial dependencies mischaracterizes epistemic uncertainty into aleatory variability. Here we introduce QuakeFormer, a unified deep learning architecture that combines these three tasks in one framework. We design a multi-station-based Transformer architecture and a flexible masking strategy for training QuakeFormer. This data-driven approach enables the model to learn spatial ground motion dependencies directly from real seismic recordings, incorporating location embeddings that include both absolute and relative spatial coordinates. The results indicate that our model outperforms state-of-the-art ground motion prediction models across all three tasks in our research areas. We also find that pretraining a uniform forecasting and interpolation model enhances the performance on early warning task. QuakeFormer offers a flexible approach to directly learning and modeling ground motion, providing valuable insights and applications for both earthquake science and engineering.

physics.geo-ph↗

arXiv2024

ChatHouseDiffusion: Prompt-Guided Generation and Editing of Floor Plans

The generation and editing of floor plans are critical in architectural planning, requiring a high degree of flexibility and efficiency. Existing methods demand extensive input information and lack the capability for interactive adaptation to user modifications. This paper introduces ChatHouseDiffusion, which leverages large language models (LLMs) to interpret natural language input, employs graphormer to encode topological relationships, and uses diffusion models to flexibly generate and edit floor plans. This approach allows iterative design adjustments based on user ideas, significantly enhancing design efficiency. Compared to existing models, ChatHouseDiffusion achieves higher Intersection over Union (IoU) scores, permitting precise, localized adjustments without the need for complete redesigns, thus offering greater practicality. Experiments demonstrate that our model not only strictly adheres to user specifications but also facilitates a more intuitive design process through its interactive capabilities.

cs.HC↗

arXiv2023

Hysteretic Behavior Simulation Based on Pyramid Neural Network:Principle, Network Architecture, Case Study and Explanation

An accurate and efficient simulation of the hysteretic behavior of materials and components is essential for structural analysis. The surrogate model based on neural networks shows significant potential in balancing efficiency and accuracy. However, its serial information flow and prediction based on single-level features adversely affect the network performance. Therefore, a weighted stacked pyramid neural network architecture is proposed herein. This network establishes a pyramid architecture by introducing multi-level shortcuts to integrate features directly in the output module. In addition, a weighted stacked strategy is proposed to enhance the conventional feature fusion method. Subsequently, the redesigned architectures are compared with other commonly used network architectures. Results show that the redesigned architectures outperform the alternatives in 87.5% of cases. Meanwhile, the long and short-term memory abilities of different basic network architectures are analyzed through a specially designed experiment, which could provide valuable suggestions for network selection.

cs.LG↗

arXiv2023

Iterative self-transfer learning: A general methodology for response time-history prediction based on small dataset

There are numerous advantages of deep neural network surrogate modeling for response time-history prediction. However, due to the high cost of refined numerical simulations and actual experiments, the lack of data has become an unavoidable bottleneck in practical applications. An iterative self-transfer learningmethod for training neural networks based on small datasets is proposed in this study. A new mapping-based transfer learning network, named as deep adaptation network with three branches for regression (DAN-TR), is proposed. A general iterative network training strategy is developed by coupling DAN-TR and the pseudo-label strategy, and the establishment of corresponding datasets is also discussed. Finally, a complex component is selected as a case study. The results show that the proposed method can improve the model performance by near an order of magnitude on small datasets without the need of external labeled samples,well behaved pre-trainedmodels, additional artificial labeling, and complex physical/mathematical analysis.

cs.LG↗