Haiwen Xu

A significant public health concern, osteoporosis is a common skeletal condition marked by decreased bone mineral density (BMD) and increased fracture risk. Conventional diagnostical methods, including dual-energy X-ray absorptiometry (DXA), present inherent shortcomings in early detection and risk stratification. Breakthroughs in machine learning (ML) and deep learning (DL) in recent years have provided novel solutions to improve diagnostic accuracy and risk prediction, and efficiency of fracture detection. This review summarizes a integrated application of ML and DL approaches in osteoporosis research, highlighting their roles in predicting BMD, screening for osteoporosis, detecting fractures, and analyzing medical imaging data from CT, MRI, and X-ray. Studies have demonstrated that ML methods, such as LR and SVM, can accurately stratify individuals at risk of osteoporosis. Meanwhile, DL models, especially CNN and transfer learning techniques, have profoundly revolutionized the detection and assessment of osteoporotic changes in medical images. The integration of these technologies is poised to transform osteoporosis management, offering more personalized and effective patient care strategies.

Taking into account the irrational elements and regret aversion of decision makers (DMs) during the decision-making process, regret theory (RT) and the TODIM methods have been integrated into a decision-making framework to develop an enhanced multi-attribute decision-making (MADM) method (PDHL-RT-TODIM) within probabilistic double hierarchy linguistic (PDHL) environment. Specifically, extending the perceived utility function in RT to determine the regret and joy values of the overall advantage flow of alternatives calculated by TODIM method in PDHL environment. Then, a correlation coefficient (CC) and standard deviation (SD) integral (CCSD) method was created using the probabilistic double hierarchy linguistic set (PDHLTS) distance metric and PDHL weight arithmetic operator to establish the objective weights of attributes. Additionally, the effectiveness of this proposed method was illustrated through numerical examples for information system investment project selection, and its stability, efficiency, and benefits were further confirmed through sensitivity analysis and comparisons with existing methods.

As flight environments and training populations grow more complex, traditional equal-weighted cumulative assessments struggle to capture the heterogeneous contribution of each checkpoint to trainee proficiency. This can suppress high performers and inflate low performers, weakening instructional improvement and risk warning. We propose a multi-dimensional dynamic weighting scheme that integrates Phi correlation, category discriminatory power (CDP), and dimension-level priority (β). Concretely, Phi (via chi-square) quantifies the association between each checkpoint and the final A/B grade; CDP describes pass-rate disparity between categories; and expert-defined β encodes dimension priorities, yielding checkpoint weights after within-dimension normalization and inter-dimension aggregation. On a 13-hour development cohort (175 trainees, 158 checkpoints), the proposed method raised Accuracy/Precision/Recall/F1 from 0.90/0.98/0.91/0.94 (baseline) to near-perfect values under the fixed institutional threshold, with clearly widened between-class margins and reduced within-class dispersion in score distributions. In a freeze-and-apply external test on a 9-hour cohort (179 trainees; one dimension absent; 50 fewer checkpoints), the method substantially outperformed the baseline (e.g., Accuracy 0.9162; F1 0.9563) without re-estimating parameters. These results suggest that dynamic weighting improves separability and interpretability and offers practical decision support for remediation, risk alerts, and promotion reviews, while remaining lightweight and deployable, offering richer data support for full-cycle training and personalized interventions.

Large-scale Aircraft Routing Problems (ARPs) remain challenging for standard Branch-and-Bound (B&B) and modern Mixed-Integer Linear Programming (MILP) solvers due to vast search spaces and instance-agnostic heuristics. Methods: We develop a learning-to-accelerate framework centered on a Two-Stage Route Selection Graph Convolutional Network (TRS-GCN) that predicts the importance of flight string variables using structural, LP relaxation, and operational features. Predictions are injected into the solver via three mechanisms: an ML-guided feasibility pump for warm starts, static problem reduction through predictive pruning, and a dynamic hybrid branching rule that blends ML scores with pseudo-costs. A synthetic generator produces realistic ARP instances with seed solutions for robust training. Results: On large instances derived from Bureau of Transportation Statistics data, TRS-GCN-guided static reduction safely pruned up to 49.2% of variables and reduced the time to reach the baseline solver’s 12-h target objective by 52.4%. The dynamic search strategy also yielded more incumbents within fixed time budgets compared with baselines. Conclusion: Integrating TRS-GCN into MILP workflows improves search efficiency for ARPs, offering complementary gains from warm-starting, pruning, and branching without changing the underlying optimality guarantees.

With the rapid advancement of the civil aviation sector and the concurrent expansion of pilot training programs, a pressing need arises for more efficient assessment methodologies during the pilot training process. Traditional written evaluations conducted by flight instructors are often marred by subjectivity and inefficiency, rendering them inadequate to satisfy the stringent demands of Competency-Based Training and Assessment (CBTA) frameworks. To address this challenge, this study presents a novel multi-label classification model that seamlessly integrates RoBERTa, a robust language model, with Graph Convolutional Networks (GCNs). By simultaneously modeling text features and label interdependencies, this model enables the automated, multi-dimensional classification of instructor evaluations. It incorporates a dynamic weight fusion strategy, which intelligently adjusts the output weights of RoBERTa and GCNs based on label correlations. Additionally, it introduces a label co-occurrence graph convolution layer, designed to capture intricate higher-order dependencies among labels. This study is based on a real-world dataset comprising 1078 evaluations and 158 labels, covering six major dimensions, including operational capabilities and communication skills. To provide context for the improvement, the proposed RoBERTa + GCN model is compared with key baseline models, such as BERT and LSTM. The results show that the RoBERTa + GCN model achieves an F1 score of 0.9737, representing an average improvement of 4.73% over these traditional methods. This approach enhances the consistency and efficiency of flight training assessments and provides new insights into integrating natural language processing and graph neural networks, demonstrating broad application prospects.

This paper focuses on the seru scheduling problem, where each seru is equipped with one work team, and the processing durations for the same type of product follow a common probability distribution. Given both subjective uncertainties (e.g., divergent expert judgments) and objective uncertainties (e.g., differences in team skill levels across serus, machine instabilities, and environmental fluctuations), processing durations are modeled as fuzzy random variables. The problem is then formulated as a mixed-integer nonlinear programming (MINLP) model to minimize the makespan of the seru production system. To facilitate computation, fuzzy random variables are transformed into trapezoidal fuzzy variables, which are subsequently defuzzified by an optimistic-pessimistic adjustment index. After obtaining the equivalent crisp model, a refined variable neighborhood search algorithm combined with simulated annealing (RVNS-SA) is designed. Extensive computational experiments are conducted on both small- and large-scale instances. For small cases, RVNS-SA achieves near-optimal solutions compared with the Gurobi solver while significantly reducing computational time. For larger cases, RVNS-SA is benchmarked against seven neighborhood-based variants and four classical metaheuristics, namely Genetic Algorithm, Particle Swarm Optimization, Artificial Bee Colony, and Tabu Search. Results demonstrate that RVNS-SA consistently outperforms the alternatives in terms of solution quality, robustness, and scalability, confirming its practical value for complex seru scheduling under uncertainty.