technical expertise

In materials science, data-scarce few-shot inverse problems—such as reconstructing microstructures of porous materials from stress–strain curves—pose significant challenges due to limited training samples and ill-posedness. Method: This work proposes a physics-informed modeling framework that systematically integrates mechanistic prior knowledge into both model architecture and training. We embed physical constraints derived from continuum mechanics and constitutive behavior to regularize learning across five distinct models: CNN, LSTM, XGBoost, Random Forest, and KNN. Contribution/Results: We conduct the first systematic evaluation demonstrating that domain knowledge consistently enhances performance across all model classes in few-shot inverse settings. Embedding physics improves feature representation, accelerates training convergence, and boosts R² for every model. Crucially, physics-informed models recover physically meaningful response patterns—e.g., strain localization and nonlinear hardening—that purely data-driven counterparts fail to capture. The framework establishes a generalizable, knowledge-augmented paradigm for solving data-limited inverse problems in materials science.

This study addresses the fundamental trade-off between accuracy and resource efficiency in deep learning systems, as well as the lack of a quality-aware optimization termination mechanism. Methodologically, it proposes a quality-driven modeling framework that systematically integrates MLOps engineering practices—including MLflow for experiment tracking and Prometheus for real-time monitoring—with domain-informed feature engineering, rule-based constraints, and model architecture customization. It further establishes a quantifiable, interpretable quality assessment system and a principled “stop-optimization” decision criterion. Contributions include: (1) significantly improved model deployment reliability and computational resource utilization; (2) a 40% reduction in quality attribution analysis cycle time; and (3) a reusable, reproducible, industrial-grade paradigm for deep learning quality governance.

Large language models (LLMs) struggle to adapt to dynamic knowledge evolution due to their static parameterization; existing knowledge injection methods predominantly focus on memory and retrieval, lacking systematic modeling and evaluation of higher-order capabilities—namely reasoning and association. Method: We propose a four-layer knowledge injection framework—Memory → Retrieval → Reasoning → Association—formally defining hierarchical progression and establishing its mapping to injection depth. We further introduce DeepKnowledge, a synthetic benchmark enabling fine-grained, depth-aware evaluation across three knowledge evolution types: novel, incremental, and updated. Contribution/Results: Through layered modeling and multi-scenario experiments, we demonstrate that advancing to reasoning and association layers significantly enhances LLMs’ dynamic knowledge adaptation. Our framework provides principled guidance for method selection across diverse knowledge evolution scenarios, bridging a critical gap between knowledge injection and high-level cognitive capabilities.

In knowledge distillation, the unavailability of the teacher model’s original training data—due to constraints such as continual learning or data privacy—poses a critical practical bottleneck. Method: This paper systematically investigates the efficacy of substitute datasets for data-free distillation. It proposes and validates non-natural images (e.g., StyleGAN-generated samples) as effective distillation sources, challenging the conventional assumption that original data is indispensable. A multi-dimensional evaluation framework is introduced to quantify distillation data quality along axes of diversity, discriminability, and feature alignment with the teacher. Contribution/Results: Through cross-domain data assessment, teacher–student feature alignment analysis, and ablation studies, the work demonstrates that diverse real and synthetic substitutes achieve distillation performance on par with original data on benchmarks like CIFAR-100—yielding up to a 3.2% accuracy gain in student models. This establishes a novel paradigm and practical guidelines for data-free knowledge distillation.

Knowledge engineering (KE) faces significant challenges in constructing large-scale, dynamic, multilingual, and multimodal knowledge graphs (KGs). Method: This study employs a hackathon-style mixed-methods approach—including interviews, ethnographic observation, artifact analysis, and empirical LLM experiments—to investigate how large language models (LLMs) can serve as effective collaborative assistants for knowledge engineers. Contribution/Results: We identify prompt engineering as a critical yet underappreciated core competency in KE practice. We introduce “KG Cards,” the first responsible AI framework specifically designed for KG construction, addressing ethical implementation gaps. Empirical results demonstrate that LLMs substantially improve KG construction efficiency; however, new bottlenecks emerge in trustworthiness assessment, cross-lingual alignment, and accountability governance. Collectively, this work provides empirically grounded guidelines and methodological foundations for human-AI collaboration in knowledge engineering.

This work addresses the limited performance of large language models (LLMs) in high-dimensional software engineering optimization tasks, where they often fail to surpass Bayesian optimization. For the first time, it systematically compares human- and AI-generated domain knowledge injection strategies and introduces four novel architectures: Human-feedback-informed Domain Knowledge Prompting (H-DKP), Adaptive Multi-stage Prompting (AMP), Dimension-aware Progressive Refinement (DAPR), and a hybrid approach combining statistical scouting with RAG-enhanced knowledge integration (HKMA). By leveraging a multi-stage, dimension-aware, and hybrid knowledge fusion framework, the proposed methods effectively incorporate structured domain knowledge to significantly enhance LLMs’ ability to generate high-quality initial solutions. Evaluated on the MOOT high-dimensional benchmark, the approaches markedly reduce the Chebyshev distance to the optimal solution and, according to Scott-Knott clustering, outperform existing LLM warm-start baselines.

This study addresses the challenge of insufficient personalized support in software engineering education, which hinders effective instruction in domain knowledge and modeling methodologies. To tackle this issue, the authors developed an intelligent tutoring system integrating a customized ChatGPT (GPT-3.5) with a course-specific knowledge base, deployed within a master’s-level course to support learning in cryptocurrency finance fundamentals and Domain-Driven Design (DDD). The system’s efficacy was validated through a five-dimensional evaluation—achieving 98.9% accuracy, 92.2% relevance, and 89.4% instructional value—and pre/post self-efficacy surveys, which revealed significant improvements in students’ confidence in applying domain knowledge and DDD principles. The project further distilled 17 reusable teaching practices, encompassing prompt engineering and curriculum integration strategies, offering a methodological innovation for incorporating generative AI into software engineering education.

This work addresses the lack of traceability in existing domain-specific fine-tuning approaches, which often leads to blind and inefficient data augmentation. The authors propose a “programming with data” paradigm that treats structured knowledge representations as a unified foundation for both training and evaluation, drawing an analogy to software development: training data serve as source code, model training as compilation, evaluation as unit testing, and data refinement as debugging. This framework enables precise, concept- and reasoning-chain–oriented model repair through structured knowledge extraction, test-driven data engineering, concept-level gap analysis, and diagnosis of broken reasoning chains. Validated across 16 disciplines, the approach significantly enhances model performance without compromising general capabilities, and the authors release an open-source knowledge base, evaluation suite, and training corpora to support reproducibility and further research.

This work addresses the frequent failure of large language model (LLM) agents in enterprise tasks due to a lack of domain-specific “tribal knowledge”—such as specialized terminology, workflows, and policies—and the inefficiency of conventional knowledge engineering approaches. Inspired by test-driven development, the authors propose a problem-first, failure-driven methodology that inverts traditional knowledge acquisition: when an agent fails on a real-world task, it actively requests the minimal set of knowledge required to resolve the specific issue. By integrating an entity metamodel with a semi-automated curation mechanism, this approach enables precise and efficient knowledge capture. Evaluated in a retail order fulfillment scenario, the method constructed a reusable knowledge base comprising 46 entities in just nine problem-resolution cycles, demonstrating both effectiveness and scalability.

Existing deep learning–based fault diagnosis methods suffer significant performance degradation on unseen programs, primarily due to a mismatch between conventional evaluation strategies and real-world deployment scenarios. This work introduces DynFault, a novel dataset comprising 38 real-world deep learning programs and 5,542 fault-injection traces, which for the first time systematically reveals and quantifies the performance gap—measured as a 0.190 drop in accuracy—between intra-program cross-validation and leave-one-program-out evaluation. The study identifies program-level feature structure as a key generalization bottleneck: curvature-based features prove effective for instability detection on unseen programs, whereas optimizer- and activation-related features exhibit predictive power only within seen programs, highlighting the heterogeneous generalizability of runtime features across programs.