arXiv subjects

Jinwen Li

Publications and source records attributed to Jinwen Li.

5 recordsLinked to original sources

arXiv2026

Cross-modal topology decodes battery faults from sparse voltage snapshots

Battery safety remains the primary bottleneck for mass electric vehicle (EV) adoption, yet field monitoring is hamstrung by a fundamental asymmetry: complex electrochemical faults must be diagnosed via sparse, low-frequency voltage measurements. Existing methods struggle to resolve the signal ambiguity between overlapping fault modes without hardware upgrades. Here, we demonstrate that these distinct fault fingerprints are not lost, but topologically folded within voltage snapshots. We introduce DeFault, a cross-modal diagnostic framework that mathematically unfolds one-dimensional voltage sequences into multi-dimensional phase-space topologies. DeFault employs a bidirectional cross-attention mechanism that acts as an autonomous, physics-aligned filter, explicitly decoding compounded fault modes that remain fundamentally invisible to sequence-based methods. Validated on a field dataset of 16.4 million data records from 99 in-service EVs, our method achieves an average accuracy of 0.96 and an F1 score of 0.84 for four fault types using only 500-second snapshots (spanning <100 mV). This work proves that high-fidelity, interpretable electrochemical diagnosis is achievable on legacy fleets without new sensors, providing a scalable solution for the battery safety crisis.

eess.SY↗

arXiv2026

Temperature-Aware Scheduling of LLM Inference in Large-Scale Geo-Distributed Edge Data Centers with Distributed Optimization

The environmental impact of Large Language Models (LLMs) on data centers hosting these models is becoming a significant concern. While many efforts have focused on reducing the substantial training overhead of LLMs, carbon and water consumption during the inference phase can often surpass the costs associated with their training. The cooling systems of data centers are crucial in this context, but they are frequently modeled with a location-independent efficiency term. However, their energy efficiency is highly influenced by ambient temperature, which can vary significantly across different geographical locations. Leveraging this temperature diversity can help reduce total cooling energy costs and improve the performance of edge data centers. To address these critical sustainability issues related to LLMs, this study proposes a temperature-aware approach that co-optimizes LLM energy costs, carbon emissions, time-to-first token, and water consumption. The approach employs a distributed optimization algorithm based on an alternating direction method of multipliers, aimed at enhancing the sustainability of LLM hosting across geo-distributed edge data centers in Australia. Our method demonstrates reductions in cooling energy consumption and improves overall cost efficiency for geo-distributed cloud environments.

eess.SY↗

arXiv2026

Leveraging Quantum Annealing for Large-Scale Household Energy Scheduling with Hydrogen Storage

Hydrogen integration into microgrids facilitates the absorption of intermittencies from renewable energy resources. However, significant challenges remain due to complex optimization problems, particularly in large-scale applications involving multiple fuel cells (FCs) and electrolyzers (ELs) with numerous binary decision variables. This paper presents a hierarchical quantum annealing (QA) model predictive control-based power allocation framework aimed at accelerating these optimization problems. First, in a day-ahead stage, the framework determines the startup and shutdown of the FCs and ELs. The short-term stage then refines the output power of the FCs and the hydrogen generation rate of the ELs. The feasibility is evaluated through a case study consisting of multiple households in Australia. Our findings demonstrate that while the traditional optimization approach performs satisfactorily in scenarios with a small number of households, the QA approach becomes more appropriate and effectively solves the problem within an acceptable range as the number of connected households increases.

eess.SY↗

arXiv2025

RoomEditor++: A Parameter-Sharing Diffusion Architecture for High-Fidelity Furniture Synthesis

Virtual furniture synthesis, which seamlessly integrates reference objects into indoor scenes while maintaining geometric coherence and visual realism, holds substantial promise for home design and e-commerce applications. However, this field remains underexplored due to the scarcity of reproducible benchmarks and the limitations of existing image composition methods in achieving high-fidelity furniture synthesis while preserving background integrity. To overcome these challenges, we first present RoomBench++, a comprehensive and publicly available benchmark dataset tailored for this task. It consists of 112,851 training pairs and 1,832 testing pairs drawn from both real-world indoor videos and realistic home design renderings, thereby supporting robust training and evaluation under practical conditions. Then, we propose RoomEditor++, a versatile diffusion-based architecture featuring a parameter-sharing dual diffusion backbone, which is compatible with both U-Net and DiT architectures. This design unifies the feature extraction and inpainting processes for reference and background images. Our in-depth analysis reveals that the parameter-sharing mechanism enforces aligned feature representations, facilitating precise geometric transformations, texture preservation, and seamless integration. Extensive experiments validate that RoomEditor++ is superior over state-of-the-art approaches in terms of quantitative metrics, qualitative assessments, and human preference studies, while highlighting its strong generalization to unseen indoor scenes and general scenes without task-specific fine-tuning. The dataset and source code are available at \url{https://github.com/stonecutter-21/roomeditor}.

cs.CV↗

arXiv2025

LG-CD: Enhancing Language-Guided Change Detection through SAM2 Adaptation

Remote Sensing Change Detection (RSCD) typically identifies changes in land cover or surface conditions by analyzing multi-temporal images. Currently, most deep learning-based methods primarily focus on learning unimodal visual information, while neglecting the rich semantic information provided by multimodal data such as text. To address this limitation, we propose a novel Language-Guided Change Detection model (LG-CD). This model leverages natural language prompts to direct the network's attention to regions of interest, significantly improving the accuracy and robustness of change detection. Specifically, LG-CD utilizes a visual foundational model (SAM2) as a feature extractor to capture multi-scale pyramid features from high-resolution to low-resolution across bi-temporal remote sensing images. Subsequently, multi-layer adapters are employed to fine-tune the model for downstream tasks, ensuring its effectiveness in remote sensing change detection. Additionally, we design a Text Fusion Attention Module (TFAM) to align visual and textual information, enabling the model to focus on target change regions using text prompts. Finally, a Vision-Semantic Fusion Decoder (V-SFD) is implemented, which deeply integrates visual and semantic information through a cross-attention mechanism to produce highly accurate change detection masks. Our experiments on three datasets (LEVIR-CD, WHU-CD, and SYSU-CD) demonstrate that LG-CD consistently outperforms state-of-the-art change detection methods. Furthermore, our approach provides new insights into achieving generalized change detection by leveraging multimodal information.

cs.CV↗