Guidelines for professional degree theses require automatically generated feedback that is both linguistically accurate and consistent with program-specific structural, methodological, and regulatory requirements. This study developed a retrieval-augmented educational artificial intelligence system that assigns confidence scores to four thesis types, retrieves type-specific normative requirements, and applies a confidence boundary mechanism to select single-type or dual-type feedback. A controlled pilot experiment compared four system variants on 50 expert-annotated Education Master’s theses: a baseline large language model, a prompt-enhanced variant, a retrieval-augmented variant, and the full system that adds the confidence boundary mechanism to retrieval. Under a strict unique-primary-label rule, the baseline and the prompt-enhanced variant reached 64.0 % and 46.0 % accuracy, respectively. The independently executed retrieval-augmented variant and the full system reached 80.0 % and 82.0 %, and this paired difference was not statistically significant. A routing-only paired reanalysis that reused the same confidence scores returned 82.0 % accuracy with the boundary mechanism both disabled and enabled, confirming that the mechanism governs feedback routing rather than the choice of the top-ranked thesis type. With the mechanism enabled, boundary detection reached 90.0 % precision, 87.1 % recall, and an F1-score of 88.5 %. Complete dual-type coverage was 61.3 %, and partial candidate coverage reached 87.1 %. Stratified-bootstrap confidence intervals and macro-averaged scores were reported to reflect the small, imbalanced sample. The results indicate that retrieval grounding supports the assignment of a primary thesis type, whereas boundary-aware routing reduces overconfident single-type guidance for ambiguous thesis topics.
Edge cloud collaborative systems face critical challenges in deploying virtual network functions with dynamic adaptability and cost efficiency due to the heterogeneity of Virtual Machine-Virtual Network Function (VM-VNF) and Container Technology-Virtual Network Function (CT-VNF) resources and high environmental dynamics. This study proposes a Long Short-Term Memory Network (LSTM)-enhanced Actor-Critic framework for VNF deployment optimization. Experimental results demonstrate that the framework achieves a deployment success rate of 97.9 %, mapping success rate of 94.8 %, average latency of 362 ms, and optimal solution fitting degree of 0.93, significantly outperforming baseline methods. With average costs 30–40 % lower than comparison frameworks under varying request loads, this study provides an effective solution for service providers to implement cost-minimization management strategies in edge cloud environments. The innovation of this study lies in designing a temporally coherent dual-network architecture rather than a simple combination of existing modules. In the actor module, LSTM encodes sequential request traces to enable anticipatory deployment decisions. In the critic module, LSTM-empowered approximation is coupled with Lagrangian relaxation to evaluate long-term constraint satisfaction under mixed VM-CT deployment. This design mitigates the high variance and slow convergence inherent in conventional Actor-Critic methods under dynamic network conditions.
Highway construction projects, characterized by large investments, extended durations, and scattered work sites, require accurate and dynamic progress monitoring to ensure cost control and safe execution. Traditional manual inspection methods suffer from inefficiency, data latency, and subjectivity, making them inadequate for modern intelligent transportation management. To address these limitations, this study proposes a novel monitoring method integrating unmanned aerial vehicle (UAV) remote sensing imagery with feature-space matching. The proposed framework involves collecting sequential site images via UAV oblique photography, generating high-precision 3D reality models through aerial triangulation and 3D reconstruction, determining optimal image segmentation scales using a multi-scale segmentation algorithm combined with the ESP2 tool, extracting construction targets via color-feature threshold segmentation and registration with BIM-based virtual models, and quantifying progress deviations through Boolean operations between 3D reality models and BIM envelopes. The primary objective is to establish a closed-loop workflow from data acquisition to progress quantification, enhancing monitoring accuracy and operational efficiency. Experimental results on a 30 km highway section over 42 days show a construction progress deviation of only 2.2 %, significantly lower than conventional methods. The proposed method provides an objective, quantitative, and visual monitoring solution, demonstrating high accuracy, stability, and practical potential for improving schedule management in highway construction projects.
This study investigates the influence of input window size and ensemble size on the forecasting performance of Decision Tree-based ensemble methods in one-step-ahead financial return prediction. Nineteen financial instruments spanning stocks, indices, commodities, and cryptocurrencies are evaluated under a chronological train-validation-test framework. Daily closing prices are transformed into log returns, and models are trained across rolling windows of 5, 10, 20, and 60 observations. The methods compared include Naive forecasting, Linear Regression, a standalone Decision Tree, Bagging-DT, Boosting-DT, and Stacking-DT, with ensemble configurations selected based on validation MAE. Linear Regression achieves the lowest average magnitude error and the best overall rank, while Bagging-DT and Boosting-DT substantially outperform both Naive forecasting and the standalone Decision Tree. The validation-selected ensemble reduces MAE across all 19 instruments, yielding an average out-of-sample improvement of 31.89 % relative to the single tree. A test-best diagnostic comparison suggests a higher improvement ceiling of 35.10 %, though this figure serves solely as an upper-bound reference. A 20-day window delivers the strongest average error performance, while a 60-day window and 100-estimator ensembles are most frequently selected at the individual instrument level. Larger ensembles incur additional computational cost without producing consistent accuracy gains. Friedman and Wilcoxon tests confirm significant performance differences among models, and Diebold-Mariano tests indicate that ensemble methods consistently outperform both Naive forecasting and the standalone Decision Tree. Overall, Decision Tree ensembles offer robust improvements over persistence-based and single-tree forecasting, though their advantage relative to Linear Regression remains instrument-dependent.
This study solves the low efficiency of green building pipeline pumps by introducing a multi-objective particle swarm optimization algorithm that integrates classification strategy and internal flow analysis. Enhanced through neural networks and update strategies, the algorithm demonstrates superior search accuracy, effectively converging to true Pareto fronts in test functions. Internal flow analysis revealed consistent velocity distribution patterns: under design and large-flow conditions, the fluid accelerated gradually from the inlet then decelerated sharply at cx/cl = 0.7. Under low-flow conditions, it showed rapid acceleration beyond relative position 0.4 with similar deceleration at 0.7. The approximate model achieved high precision, with maximum prediction errors of 1.96 %, 2.37 %, and 0.42 % across different flow rates. The characteristic of optimized design was the extended straight inlet section, which produced a more uniform velocity distribution while maintaining cross-sectional similarity with the original model at a relative position below 0.5. This methodology provides an effective approach for structural optimization of energy-efficient pipeline pumps in green buildings.
The growth of digital business and communication systems has brought about an increase in risks. As the digital world expands, so does the use of its components. Trillions of transactions occur online every hour, making safety crucial when exchanging data over a public communication channel. Cryptography is one method for providing this safety. While many cryptographic algorithms have been developed using graph theory and conventional mathematics, few articles on cryptography have been published that use artificial neural networks and DNA structure as their basis. The researchers attempted to discover new methods with the latest technology to keep confidential information secure and improve the robustness of the communication system. In this paper, two cryptography algorithms have been presented that are built on the idea of tree parity machines. The input message is converted to binary and fed into the encryption system. The key has been generated dynamically using the ANN technique. The performance of the developed algorithms has been evaluated and analyzed considering the parameters such as execution time, throughput, memory use, and strength. The obtained outcomes were found to be comparatively better than existing algorithms.
In basketball motion recognition, there are problems such as difficulty in obtaining high-quality action data, difficulty in coping with poor discrimination accuracy due to scene factors such as movement of athletes, occlusion of equipment, and interference of spectators. Aiming at these problems, the study proposes to use density-based spatial clustering of applications with noise to construct a basketball motion recognition model. On the basis of this model, a joint visual-inertial sensor is introduced to collect the action data of basketball, and a hybrid Gaussian model and dynamic time warping algorithm are added to realize the recognition of single joints and multi-joints. The outcomes found that on the performance test, the proposed model of the study had the best overall performance with a loss rate of 1.02 %, an accuracy rate of 97.02 %, a recall rate of 95.26 %, an F1 value of 95.17 %, a discrimination time of 1.3 s, and an error rate of 0.9 %. In addition, in terms of motion trajectory quality, the joint vision-inertial sensor designed by the study had better quality of captured motion trajectory smoothness than the single type of sensor. In the example test, the proposed model could well recognize six basketball movements: dribbling ball, passing the ball, shooting, side step defense, crossing over, and dealing. The correct identification rate was no less than 90 %, and the recognition time was no more than 2 s. The clustering of features also showed clear clusters, and the model was able to correctly recognize the complex basketball scenarios. Taken together, the research model can be well applied to basketball motion recognition and promote the improvement of basketball players’ action level.
Building reliable automatic speech recognition (ASR) systems for low-resource languages such as Kannada is challenging due to the scarcity of training data and the cross-lingual translation requirements. This study investigates fine-tuning Whisper models – Tiny, Small, and Medium – to achieve end-to-end Kannada-to-English speech transcription. Using a dataset of Kannada speech, the multilingual Whisper models are fine-tuned to improve transcription and translation performance. The accuracy of the fine-tuned models was evaluated using word error rate (WER) and character error rate (CER), revealing clear improvements across the different model sizes. Based on the analysis of the experimental results, the Whisper-Medium model was found to perform the best, with significant reductions in WER and CER compared to the zero-shot baselines. The results highlight the feasibility of lightweight and midsize Whisper architectures for cross-lingual ASR in low-resource Indian languages and indicate that these models are more suitable for real-world applications in multilingual speech systems.
To address issues such as information loss and insufficient restoration accuracy in the process of cultural relic protection and restoration, this study proposes a digital protection and virtual restoration method for cultural relics that integrates multimodal data, aiming to enhance the intelligence and refinement level of cultural relic restoration. This method comprehensively utilizes multimodal data including pattern textures, structural lines, coloring, and material properties, and conducts virtual restoration based on the Gated Convolutional Generative Adversarial Network (GC-GAN) framework. The method reconstructs the structural lines of cultural relic patterns through adaptive curve fitting technology, and combines material attribute information to achieve accurate restoration of damaged areas, thereby improving visual consistency while maintaining historical authenticity. The GC-GAN consists of a coarse generator, a fine generator, and a discriminator, which effectively restore damaged images. Experiments show that GC-GAN outperforms traditional algorithms like Criminisi, especially in SSIM and PSNR metrics. The average improvements are 7.4 % and 30.5 %, respectively. In the overall validation of the virtual restoration method, this approach achieves better restoration effects and image quality than other methods. The average SSIM and PSNR improvements are 0.83 % and 6.19 %, respectively, with stable restoration results. Although the runtime is slightly longer than other methods, it still meets practical application needs. Tests with different missing rates demonstrate strong robustness at lower missing rates. The research results provide a feasible technical path and new ideas for the digital protection and virtual restoration of cultural relics, and possess significant practical application value and promotion potential.
Turing models of pattern formation provide insight into an intriguing question in developmental biology, like how nature exhibits various structures, shapes, and organized patterns. These natural patterns include the pattern and texture on a desert dune, spots and stripes on the skin of various animals, the growth of a body from a single cell, and the formation of fingerprints. The current work emphasized stability analysis and parameter settings to obtain diverse patterns in nature. As growth is an inevitable continuous process, it is responsible for producing different structures and patterns in living beings. The proposed numerical framework and simulation exhibit realistic natural patterns using reaction-diffusion (RD) models driven by Turing-type instability in the Gray-Scott model. The study proposes a parameter space for Turing patterns using stability analysis. The implemented mathematical model discretizes the continuous partial differential equations into their discrete counterpart by employing a finite-difference scheme. The presented framework combines state-of-the-art spatial and temporal discretization techniques together with stability analysis to mirror stable Turing-type patterns. The proposed numerical scheme is robust, efficient, accurate, and capable of exhibiting diverse biological patterns for the set of parameters in the Turing space, which are validated through stability analysis.
In the rapidly developing digital media era of artificial intelligence, image editing technology has become the core support for digital content creation and information transmission. However, existing image editing technologies still face key challenges such as unstable generation quality, low interaction efficiency, and insufficient cross domain adaptability. To address these issues, a digital media image editing algorithm based on image generation and manual interaction is proposed, aiming to achieve the unity of high-quality image generation and precise control by integrating hybrid generation architecture and dynamic attention mechanism. Experiments show that the research algorithm achieves a Fr & eacute;chet Distance (FID) value of 21.3 in terms of generation quality, which is 44.8 % and 53.9 % higher than the control group's values of 38.8 and 46.2, respectively. In the cross domain editing task, when the intersection to union ratio index reaches 89 %, the Dynamic Similarity Adjacency Matrix (DSAM) value of the research algorithm is 0.83, which is 21.4 % and 18.7 % higher than the control group method, respectively. In terms of system stability, the research algorithm has a misidentification rate of only 6.0 % and a crash frequency of 1 after 5 h of continuous operation, which is much better than the control group's 34.5 % and 5 crashes. This proves that the research algorithm can effectively balance generation quality and computational efficiency, improve user interaction experience while maintaining structural consistency, and provide technical reference for professional level image editing.
In the construction of smart cities, logistics warehouse robots have become a key tool for improving logistics and warehousing management efficiency due to their efficient and accurate characteristics. To further optimize the logistics workflow and reduce collision and waiting time, a new trajectory prediction model is proposed. The Stanley algorithm and model predictive control module are introduced into the trajectory prediction model, which can help the model predict the direction and distance of the automated guided vehicle in real time, and correct the prediction results. The research results indicated that the trajectory prediction model had good adaptability and accuracy, which accurately predicted various types of motion trajectories. The average deviation of the trajectory prediction model was only 8.65 %, the lowest tracking error was 2.35 m, and the average computation time was 14.15 ms. The trajectory prediction model improved the accuracy of path prediction by 16.05 % compared with traditional long short-term memory network algorithms, with a precision of 0.94. From this, it can be seen that the trajectory prediction model can accurately predict the motion trajectory of automatic guided vehicles. The model proposed by the research institute provides a new tool for path planning of logistics robots, which is helpful for the construction of smart cities. In terms of logistics management, it helps to improve the management efficiency of staff. Management personnel can timely understand the operation status of logistics operations based on the data provided by the model, and make more scientific and reasonable management decisions. The new model can monitor the status of robots in real time, detect wear and tear or abnormalities in advance, and avoid the backlog of goods and interruption of operations caused by equipment failures.
The freezing of gait (FoG) presents a sudden challenge in sustaining movement which becomes a common gait issue in people with later stages of Parkinson’s disease (PD). FoG often results in falls that reduces the individual’s impact on life. A highly precise detection technique is required for accurate detection of FoG episodes automatically. This paper utilizes multivariate signal decomposition techniques, including Variational Mode Decomposition (VMD), Multivariate Variational Mode Decomposition (MVMD), and Successive Variational Mode Decomposition (SVMD). These techniques are utilized to extract time-frequency domain features from FoG signals. The Daphnet FoG dataset is used to evaluate performance in the studies. The features derived from the decomposition techniques serve as inputs to classifiers. In this study, five classifiers are employed including both machine learning and deep learning methods. The study attained the highest classification accuracy of 96.74 % with the use of a 1D CNN. The proposed approach demonstrates the potential for advancing the automated detection and facilitating early-stage diagnosis and intervention in Freezing of gait.
Multiple-Object Tracking (MOT) is a fundamental task in computer vision with many applications. For practical operations, tracking for monitoring with thermal imaging unaffected by lighting conditions is important. However, most MOT methods are proposed to analyze video streams from RGB cameras, while there are few datasets and research on multi-object tracking in infrared image sequences. In this paper, we provide a new infrared dataset for object detection and tracking, which contains small objects and occlusion challenges. We also propose a new robust tracker, which enhances object detection with the strategic integration of the Convolutional Block Attention Module (CBAM) into the YOLOv7 model, along with specialized fusion of IoU, Size, and ReID features during data association to overcome the challenges of thermal images. Our tracker achieves 59.29 HOTA, 73.46 MOTA, and 74.4 IDF1 as a new state-of-the-art on the CAMEL benchmark. The tracker's source code and dataset are publicly available at: https://github.com/aquarter147/TMTV_Thermal_MOT
In the globalized and fiercely competitive market environment, supply chain management plays a pivotal role in the operation and development of enterprises. A supply chain management optimization model based on improved genetic algorithm is proposed. Through simulation experiments, the performance in time, cost, and efficiency is verified. It is also compared with traditional strategies. The experimental results indicate that most of the time changes in the simulation process and standard delivery are roughly the same. The subtle difference appears after 600 s. There is a significant difference at 1,100 s, but it is quickly self calibrate. The inventory change value based on the inventory turnover control strategy is 0.2547, while the inventory change value based on the improved genetic algorithm control strategy is 0.2734, with a difference of 5.9 %. The majority of supply chain efficiency under two control strategies is located in high efficiency areas, with efficiency greater than 92 %. When delivering, in addition to supply efficiency, the efficiency of turnover days also needs to be considered. Most of the loss rate points are within 220 g/km. Compared to traditional control strategies, control strategies based on improved genetic algorithms have some advantages in the distribution of loss rates, with more points distributed between 190 and 210 g/km. Furthermore, compared with traditional methods, the proposed model has advantages in cost, time, and efficiency. Therefore, this research provides an optimization strategy for supply chain management, which is beneficial for improving the operational efficiency and development potential of enterprises in a globalized and fiercely competitive market environment.
This paper presents the results of an empirical study investigating the impact of unit test quality on the development time, the ability to solve all tasks, the perceived software understanding, and the presence of adequate feedback. For this purpose, an application was developed with two test suites, one containing high-quality tests and the other low-quality tests. The study was performed on 53 graduate students of computer science, who were randomly divided into three groups: students with high-quality tests (Q), students with low-quality tests (L), and students with no tests (N). The goal was to observe the differences between the groups when solving different tasks. Based on the results obtained, it can be said that a significant time difference could be discovered between the groups Q and L, as the Q group completed the individual tasks in a shorter time. In addition, a connection between the presence of quality tests and having adequate feedback has been found, though there was no significant difference in perceived code understanding or ability to solve all tasks.
In order to improve the analysis effect of the international competitiveness of China’s new energy vehicle industry, this paper combines the improved AHP and grey relational analysis method to analyze the international competitiveness of China’s new energy vehicle industry. This paper uses the quantitative information analysis algorithm of new energy vehicle competitiveness to process the signal data of China’s new energy vehicle industry’s international competitiveness. Moreover, this paper collects data through various channels, and applies the improved AHP and grey relational analysis to the analysis of the international competitiveness of China’s new energy vehicle industry. In addition, this paper proposes an improved SP & SPWVD combined time-frequency distribution algorithm. It combines the long-window short-time Fourier transform (STFT) with better frequency resolution and the short-window STFT with better time resolution to obtain a combined window spectrogram with better time-frequency aggregation. The research results show that when the improved AHP and grey relational analysis method are used in the analysis of the international competitiveness of China’s new energy vehicle industry, the analysis effect of the international competitiveness of China’s new energy vehicle industry can be effectively improved.
To solve the insufficient educational resources in offline classroom English teaching, the research focuses on online teaching and designs an online grammar correction algorithm to help students realize automatic online error correction. The algorithm is based on the Transformer model algorithm based on multi-head attention mechanism (MHAM), and integrates word order information into the encoding process. Three pseudo-parallel corpora are used to expand the number of training data. Finally, Adam is used to optimize the model parameters to improve the model performance. The accuracy, recall, and F0.5 of the algorithm formed after two one-way optimization are the highest values in the same type of optimization model. The SP + Prehuman + TFcopy algorithm formed after double optimization has the best comprehensive performance. The accuracy rate reached 68.53 %, and the F0.5 value reached 58.26 %, both of which were the highest values in the comparison model. Moreover, this method can also provide certain technical support for the establishment of multilingual interaction platforms and high-quality natural language generation.
This article presents L-NeRVEn, a software solution for after-stroke upper limb neurorehabilitation utilizing collaborative virtual reality (VR). L-NeRVEn offers a shared virtual environment, experienced via a VR headset, with animated full-body avatars representing both the therapist and the patient. The patient’s task is to imagine a particular movement with his or her disabled limb and the success of this imagination is determined by the processing and classification of the patient’s electroencephalography (EEG) signal. The EEG processing component is loosely coupled with the rest of the solution and can be easily modified or replaced. The article focuses on the architecture and appearance of L-NeRVEn, explains the role of its users, and the interaction of its components. It also presents the results of L-NeRVEn evaluation with n = 18 participants, utilizing subjective (SUS, IPQ and NASA-TLX questionnaires) and objective (video recordings and observer’s notes) methods.
With the continuous evolution of smart environments powered by Internet of Things (IoT) networks and smart devices, there becomes a crucial need to address and ensure privacy and security. Intrusion Detection Systems (IDSs) that are specially designed for use in IoT networks play a vital role in strengthening the security posture of an IoT network and system by safeguarding and preventing attacks against smart environments. This research paper presents a comparative study of IDSs for IoT networks, with a focus on signature-based, anomaly-based, and specification-based IDS detection methods while highlighting the significance of IDSs in protecting IoT networks and smart environments, which have become recent targets for attackers due to their integration with modern and advanced technologies and their involvement with large volumes of data. The study investigates the mentioned IDS methods covering the strengths and weaknesses of each method in safeguarding smart environments and networks. This paper dives into the characteristics that make IDS decision-making more effective primarily in terms of security, considering privacy and performance. The findings of this study contribute to the hardening of IoT network security by offering recommendations for IDS selection for enhancing IoT overall security, specifically through the adoption of adaptive-based IDSs.
