Selected work

Representative Papers

Artificial intelligence in materials science and engineering: Current landscape, key challenges, and future trajectories

Jul 01, 2025Composite structures

Materials development faces significant challenges including data complexity, lengthy timelines, and low efficiency, necessitating intelligent approaches to accelerate discovery. This work provides a systematic review of artificial intelligence applications in materials science, integrating a spectrum of techniques from traditional machine learning to deep learning and generative AI. It focuses on representation methods and model construction for multimodal data—such as composition, structure, images, and text—and covers core algorithms including convolutional neural networks (CNNs), graph neural networks (GNNs), Transformers, and Gaussian processes. The review particularly highlights emerging directions such as uncertainty quantification, multi-source data fusion, and language-inspired representations, proposing a research pathway toward intelligent materials design. By offering a comprehensive AI framework, this study identifies critical challenges in data quality, standardization, and algorithmic adaptability, thereby significantly enhancing the efficiency and reliability of materials discovery and optimization.

10 citationsRead paper

Survey in Characterization of Semantic Change

Feb 29, 2024arXiv.org

This paper addresses the adverse impact of semantic change on computational linguistics tasks—including machine translation and information retrieval—by systematically surveying its representation methods. Focusing on lexical meaning evolution across cultural, domain-specific, and diachronic dimensions, it proposes, for the first time, a three-dimensional formal taxonomy: *dimension* (generalization/narrowing), *polarity* (amelioration/pejoration), and *relation* (metaphor/metonymy), unifying existing paradigms while clarifying theoretical boundaries and evaluation criteria. Leveraging temporal word embedding modeling, distributional semantics, historical corpus analysis, and semantic graph techniques, the work constructs a comparative matrix of mainstream approaches. Results reveal a critical research gap: current efforts emphasize *detection* of semantic change but underprioritize *interpretability* and *mitigation*. The study thus advances the field toward explainable semantic change representation and controllable impact modeling.

10 citationsRead paper

Onboard Satellite Image Classification for Earth Observation: A Comparative Study of ViT Models

Sep 05, 2024

This work addresses land-use classification of onboard remote sensing imagery under resource-constrained, high-noise, and low-power in-orbit inference conditions. We systematically evaluate and optimize visual Transformer architectures for spaceborne deployment. For the first time, we comparatively assess ViT variants (EfficientViT-M2, MobileViTV2), CNNs, and ResNets in an Earth observation (EO) satellite environment, proposing a model selection paradigm centered on energy efficiency and noise robustness. Experiments on the UC Merced dataset show that EfficientViT-M2 achieves 98.76% classification accuracy with only 10 seconds inference time per image and 1000 seconds training time. Compared to MobileViTV2, it reduces power consumption by 63.93% and improves robustness score to 0.79—yielding the best overall performance. Our results establish EfficientViT-M2 as a new deployable, highly robust, and energy-efficient vision backbone benchmark for intelligent onboard remote sensing.

4 citations1 influentialRead paper

An Empirical Study of the Imbalance Issue in Software Vulnerability Detection

Feb 12, 2026European Symposium on Research in Computer Security

This study addresses the performance instability of deep learning models in software vulnerability detection, which primarily stems from extreme class imbalance due to the scarcity of vulnerable samples. For the first time, it systematically validates class imbalance as the core factor driving such performance fluctuations. The authors empirically evaluate the impact of prominent class-imbalance mitigation strategies—including Focal Loss, mean squared error, class-balanced loss, and random oversampling—across nine open-source datasets and two state-of-the-art deep learning models. Their findings reveal that Focal Loss significantly improves precision, mean squared error and class-balanced loss yield better recall, and random oversampling achieves the highest F1 score. However, no single method consistently outperforms others across all evaluation metrics.

4 citationsRead paper

Test smells in LLM-Generated Unit Tests

Oct 14, 2024arXiv.org

This study addresses the understudied issue of test smells in large language model (LLM)-generated unit tests. Method: We conduct the first large-scale empirical analysis of test smells across 20,505 Java class-level test suites drawn from five sources: human-written tests, EvoSuite-generated tests, and LLM-generated tests from GPT-3.5, GPT-4, Mistral, and Mixtral. Our multi-benchmark, cross-model analysis framework encompasses over 770,000 test cases, leveraging dual smell-detection tools—TsDetect and JNose—across 34,635 open-source projects and the TestBench benchmark. Contribution/Results: We identify prevalent smells—including Assertion Roulette and Magic Number Test—in LLM-generated tests; their occurrence patterns are significantly influenced by prompting strategies, context length, and model scale. Notably, LLM-generated tests exhibit smell profiles closer to human-written tests than to search-based software testing (SBST) outputs, suggesting potential training data contamination. These findings provide critical empirical grounding for developing smell-aware test generation frameworks.

4 citationsRead paper

Recent publications

Latest Papers

Evaluating OpenMP Offloading for Intra-node Multi-GPU Programming across NVIDIA, AMD, and Intel Architectures: A 3D Heat Transfer Case Study

Aug 12, 2026

This study addresses the challenge of achieving both cross-architecture portability and high performance in single-node, multi-GPU scientific computing across NVIDIA, AMD, and Intel platforms. Focusing on a 3D heat conduction problem, it presents the first unified implementation of multi-GPU collaboration using OpenMP offloading across all three major GPU architectures, and systematically evaluates its communication efficiency, memory management, and scalability relative to native programming models—namely CUDA, HIP, and SYCL. Experimental results demonstrate that OpenMP offloading achieves approximately 2× and 4× speedup on dual-GPU and quad-GPU configurations, respectively, confirming its potential to deliver near-native performance while maintaining strong portability across heterogeneous hardware ecosystems.

0 citationsRead paper

Memorization Diagnostics for Code LLMs Should be Scale-Aware

Aug 12, 2026

Existing diagnostic methods for assessing memorization in large code models struggle to disentangle memorization from representational capacity as model scale increases, leading to distorted evaluations. This work proposes a novel paradigm that decouples representational load from memorization behavior through invertible mathematical transformations, complemented by systematic analyses employing synonym obfuscation, dead code insertion, and log-probability probing techniques. The study reveals that current probing approaches significantly fail on large-scale models, while the models themselves exhibit strong robustness to diverse surface forms of code. These findings challenge the validity of relying solely on contaminated benchmarks to evaluate memorization and provide both theoretical grounding and methodological support for reassessing the generalization capabilities of large code models.

0 citationsRead paper

FlowGRN: Scalable and Dropout-Robust Gene Regulatory Network Inference via Flow Matching-Based Trajectory Reconstruction (Technical Report)

Aug 10, 2026

Inferring gene regulatory networks (GRNs) from single-cell RNA sequencing data is challenging due to the absence of temporal information and high levels of dropout noise. This work proposes a novel approach that, for the first time, integrates conditional flow matching with score matching to enable robust reconstruction of dynamic cellular trajectories. The method further incorporates a scalable dynGENIE3 framework for GRN inference and introduces a dropout-robust cell similarity metric. Evaluated on the BEELINE benchmark, the proposed method achieves state-of-the-art performance across both synthetic and real datasets. Ablation studies confirm the effectiveness of each core component, demonstrating their individual contributions to the overall improvement in GRN inference accuracy.

0 citationsRead paper

FlowGRN+: Improving Gene Regulatory Network Inference by Spline Fitting and Manifold Projection in Conditional Flow Matching (Technical Report)

Aug 10, 2026

This study addresses the challenges in inferring gene regulatory networks (GRNs) from single-cell RNA sequencing data, which arise from high dimensionality, dropout noise, and temporal discontinuities in dynamic trajectories. To overcome these issues, the authors propose a novel conditional flow matching framework that integrates spline fitting to construct a stable reference trajectory and introduces a manifold-based local tangent space projection to suppress overshooting. This approach significantly enhances temporal consistency in cellular dynamics modeling and improves GRN inference accuracy. By promoting trajectory smoothness and minimizing manual intervention, the method demonstrates superior performance and higher reproducibility on the BEELINE benchmark compared to existing approaches.

0 citationsRead paper

Metanormative Theory for RL-Based Moral Agents

Aug 08, 2026

Current reinforcement learning agents lack a unified framework for moral judgment and theoretical guidance, making it difficult to ensure their behaviors align with human values. This work introduces meta-normative theory into the reinforcement learning paradigm for the first time, establishing a design framework for moral agents that integrates philosophical foundations with computational mechanisms. It further proposes an operational taxonomy and evaluation criteria for morally relevant agent behaviors. By clearly defining standards for assessing the morality of reinforcement learning agents, this study not only advances the theoretical grounding for comparing, selecting, and refining value-alignment methods but also facilitates the transition of machine ethics from abstract principles toward quantifiable evaluation.

0 citationsRead paper