This paper presents a comprehensive survey of join order selection (JOS) in database query optimization, with a particular focus on deep reinforcement learning (DRL)-based approaches. By synthesizing representative systems such as Google Balsa and Microsoft Adaptive Query Processing, we characterize current trends and open challenges in reinforcement learning-driven query optimization. We formulate the multi-way join ordering problem as a Markov Decision Process (MDP) and systematically compare key design dimensions, including state representation (e.g., graph neural networks versus vectorized statistical features), action space design (e.g., join pair selection versus join subtree expansion), and reward function design (e.g., sparse versus intermediate rewards), analyzing their impacts on optimization outcomes. For multi-objective optimization, we survey strategies such as Pareto optimization and scalarization, highlighting their effectiveness in balancing query latency reduction with resource consumption and execution stability. To enhance model generalization, we examine the roles of curriculum learning and meta-learning in improving sample efficiency and robustness across heterogeneous workloads. In addition, we categorize hybrid architectures that integrate DRL with traditional cost-based optimizers (CBOs), including designs where DRL functions as a query rewriter, a search-space guide for plan enumeration, or a post-optimization validator. Finally, extensive empirical evaluations on standard benchmarks such as TPC-H and the Join Order Benchmark (JOB) demonstrate that DRL-based methods can outperform classical optimizers, particularly for complex queries with large search spaces.
This paper provides a systematic review and comprehensive analysis of network and information security, addressing the complex threats currently confronting cyberspace. Starting from an examination of various types of cyber attacks and their technical characteristics, and incorporating case studies of typical security incidents, this paper reveals the evolution trends in attack methodologies and their impact on critical information infrastructure. At the technical level, this study constructs an information security technology framework centered around encryption and authentication technologies, as well as intrusion detection and defense mechanisms, while exploring implementation pathways for multi-layered security protection. At the management and policy level, this study introduces a security risk assessment methodology, discusses the processes of risk identification, quantification, and hierarchical control, and proposes optimization recommendations regarding the establishment of security policies and emergency response mechanisms. This paper aims to serve as a reference for technical architecture design and security governance practices within the field of network and information security, while providing theoretical support for enhancing overall security protection capabilities and risk mitigation capabilities.
As the automotive industry transitions from "incremental expansion" to "deepening existing markets," digital marketing is emerging as a critical battleground for cost reduction and efficiency enhancement. This study focuses on the "platform AI large model foundation" system, exploring how to integrate open-source large models with domain-specific automotive datasets to train vertical models tailored for digital marketing scenarios in the automotive sector, thereby supporting core capabilities such as semantic understanding and content generation. The research constructs a specialized automotive marketing corpus covering vehicle analysis, competitive benchmarking, user profiling, and marketing messaging. It proposes a dual-model training architecture combining LoRA-based parameter-efficient fine-tuning and RAG (Retrieval-Augmented Generation), and integrates knowledge graphs into the marketing decision-making pipeline. This work provides a practical technical foundation and systematic methodology for the intelligent transformation of digital marketing in the automotive industry.
Lithology identification utilizing logging data stands as a pivotal research area within reservoir prediction, integral to the broader context of oil and gas exploration and development. In order to solve the problem that the traditional logging lithology identification method and machine learning method do not have high accuracy for the lithology identification of complex carbonate rocks. In view of the powerful performance of deep learning methods in feature extraction and data analysis, this paper proposes a lithology recognition method based on the Genetic Algorithm-adaptive LightGBM model. Due to the small number of features of the original logging data, the feature shift method is first used to expand the features of each data. Subsequently, to optimize the classification performance of our model, we leverage the Genetic Algorithm (GA) to fine-tune the hyperparameters of the LightGBM model. With the optimal hyperparameter configuration in place, we evaluate the model's performance through rigorous training with logging data. The adaptive LightGBM lithology recognition model based on GA is compared with the lithology identification results of GBDT, Bayes, DNN and DT, and the overall prediction performance is evaluated by using the accuracy and its macroscopic mean. Through the test and verification of the actual well data, the proposed method has achieved very considerable results in lithology identification, showing the broad prospect of LightGBM technology in the field of geophysics.
Review of the application of artificial intelligence (AI) in wearable health-monitoring devices and analysis of their development from basic activity monitors to clinical diagnostic instruments. Wearable sensors can collect continuous health data now; therefore, machine learning algorithms need to be introduced to process this complex, high-dimensional data. Analyze the particular algorithmic structures in wearable ecosystems for this study, including supervised learning for anomaly detection and deep learning for time-series forecasting. Review of practical clinical applications: cardiovascular continuous monitoring, management of metabolic disorders and early diagnosis of neurological diseases. In addition, this paper will also investigate the technical and regulatory problems of wearable artificial intelligence, such as data privacy, system interoperability and algorithmic bias. Optimisation strategies are proposed in the paper, such as edge computing and federated learning, to process data locally and reduce privacy risks while maintaining model accuracy. Based on the above analysis, a unified regulatory system and transparent machine-learning protocols need to be established for the incorporation of wearable AI into official medical institutions.
Fixed-interval candlestick sampling cannot adequately represent the non-uniform arrival of information in futures markets. This study develops an information-driven market-regime detection and trading framework. One-minute data are resampled along a hybrid information axis constructed from standardized trading-volume intensity, realized volatility, price momentum, and a high-low spread proxy. A GMM-HMM identifies latent regimes, after which XGBoost and CatBoost are trained on 400-dimensional lagged-price features; trades are executed only when both models agree. Using rebar futures data from 2010 to 2025 and reserving 2023-2025 for out-of-sample testing, the complete strategy achieves an annualized return of 11.33%, a maximum drawdown of -7.80%, and a Sharpe ratio of 0.88, outperforming conventional sampling, single-model, and no-regime-filtering alternatives in ablation tests. The results show that combining information-driven resampling with regime-aware model fusion improves trade selectivity and risk-adjusted performance.
Generative artificial intelligence is reshaping the psychological foundations of photography. For a long time, photography gained its persuasive force from physical presence, optical recording, and temporal trace. Viewers tended to read photographs as evidence of an encounter between a camera, a subject, and a moment. AI-generated images disturb this habit. They can look photographic without requiring a photographed object, a shared scene, or bodily participation by the maker. To analyse this shift, this study examines the transformation of photographic psychology across: media authenticity, visual judgment, algorithmic bias, copyright and authorship, memory practice, and identity construction on social platforms. This suggests that generative AI does not simply replace photography. It reorganizes photography into a hybrid visual system where recording and generation coexist. As a result, photographic psychology shifts from the psychology of capturing reality to the psychology of judging sources, negotiating trust, and constructing identity under algorithmic conditions. The continuing value of photography lies not only in visual realism, but also in human experience, ethical responsibility, situated attention, and the durable connection between images and lived reality.
Analyze the practical value and optimisation directions of sensor intelligent perception schemes based on the STM32 platform in this paper. STM32 is not only a low-cost control chip but also suitable for implementing integrated sensing and local processing, as well as communication and lightweight intelligent decision-making in embedded systems, as this paper explores. Based on smart sensor theory, Internet of Things architecture, multisensor fusion studies and recent STM32 artificial intelligence tools, this essay explores the prospects of STM32-based schemes for real-time response, deployment flexibility, system integration and edge-side data utilization. The following are the general optimisation directions of such schemes: sensor selection, interface coordination, fusion algorithms, energy management, memory optimisation, edge AI deployment and reliability enhancement. Therefore, the value of STM32-based intelligent perception lies not only in hardware cost but also in the combined design of a sensing architecture, embedded computing resources and application-specific optimisation strategies.
Village waste management currently has two main problems: lack of a digitally integrated waste recording system and low community participation in sorting waste due to lack of incentives. This study aims to design a prototype of the ECO Notes application as a village-based waste recording solution, adapting features and interfaces through a User-Centered Design (UCD) approach. Comparison with current methods shows that the Digital Waste Bank focuses more on transaction recording, while the IoT-based Waste Monitoring System, effective for city scale, is less suitable for village limitations. The design methodology includes conducting literature studies and interviews with village officials and local communities to identify functional and non-functional needs, followed by the creation of low-fidelity and high-fidelity prototypes. Evaluation was conducted using a User Experience Questionnaire (UEQ) comprising six rating scales on 12 respondents. The evaluation results showed that the highest scores in the Perspicuity aspect (1.68), indicating ease of understanding of the application, and Novelty (1.25), reflecting the solution's novelty compared to previous methods. Overall, the application received positive responses in all aspects, confirming its advantages in accessibility, sustainability, and community engagement. Thus, ECO Notes is considered worthy for further development and testing on a wider-scale user
To address the communication security of the Intelligent Reflecting Surface (IRS)-assisted Multicarrier Non-Orthogonal Multiple Access (MC-NOMA) system, a problem of maximizing the secure sum rate is formulated under the constraints of subcarrier allocation schemes, base station transmission power, and IRS phase shifts. The security performance of the communication system is enhanced by jointly alternating optimizing beamforming vectors and IRS phase shifts. Simulation results demonstrate that the proposed scheme exhibits significant advantages in improving the security and efficiency of communication systems.
This Accurate and interpretable report generation from fundus images remains a critical yet challenging task in medical artificial intelligence, particularly due to the static nature of model adaptation and the limited evolvability of existing agent-based frameworks. Although ophthalmic foundation models have significantly improved visual representation learning through large-scale self-supervised pretraining, they lack mechanisms for continual adaptation during inference. Meanwhile, current agent-based approaches enhance reasoning but remain constrained by fixed cognitive structures. In this work, we propose a self-evolving diagnostic framework that unifies parametric adaptation and cognitive evolution for fundus report generation. Specifically, we introduce a gated residual adapter to enable dynamic, inference-time knowledge integration while preserving prior knowledge. Furthermore, we develop a medical agent architecture based on the OpenClaw paradigm, which continuously refines its core reasoning strategy through physician feedback and consistency-driven constraints. By coupling model-level adaptability with agent-level reasoning evolution, the proposed framework enables sustained performance improvement in complex clinical scenarios. This work provides a new perspective on building continuously evolving intelligent diagnostic systems for real-world healthcare applications.
With the rapid development of the Internet of Things (IoT), cloud computing, and 5G communication technologies, network protocols, as the core of network communication, are facing increasingly severe security threats. Fuzz testing is an effective technology for detecting vulnerabilities in network protocols; however, traditional fuzz testing methods have limitations such as low test efficiency, blind test case generation, and difficulty in adapting to complex and diverse network protocols. In recent years, the rapid advancement of artificial intelligence (AI) technology has brought new opportunities for the innovation and development of network protocol fuzz testing. This paper systematically studies the application of AI technology in network protocol fuzz testing. First, it elaborates on the urgency of the current network security situation and the importance of network protocol fuzz testing. Then, it introduces the basic principles, core processes, and development stages of network protocol fuzz testing technology. Next, it focuses on the application methods of AI technologies such as machine learning, deep learning, and reinforcement learning in network protocol fuzz testing, and analyzes the improvement effects of AI on fuzz testing efficiency, coverage rate, and vulnerability detection accuracy. Subsequently, it explores the practical application scenarios of AI-based network protocol fuzz testing. Finally, it summarizes the current deficiencies of AI-based network protocol fuzz testing technology and looks forward to its future development trends. This study provides a theoretical reference and practical guidance for the research and application of network protocol fuzz testing technology in the AI era, helping to improve the security and reliability of network protocols.
Electroencephalography owing to its high temporal resolution and non-invasive nature, holds significant potential for auxiliary diagnosis in neuropsychiatric disorders such as schizophrenia. However, conventional EEG analysis methods often fall short in capturing high-order structural features of complex brain functional connectivity and in achieving robust generalization across subjects. To address these limitations, this study proposes a novel image classification framework, CEResNet, which integrates topological data analysis with color-equivariant convolutional neural networks. Specifically, the proposed method first employs persistent homology to extract multi-scale high-order topological features from multi-band EEG signals. These features are then transformed into persistence images via Gaussian kernel mapping, serving as inputs for the deep learning model. A color-equivariant convolutional module is subsequently embedded into the ResNet-18 architecture to enhance the model’s robustness against color distribution shifts. Experimental results demonstrate that the proposed approach achieves a classification accuracy of 0.9231 in distinguishing schizophrenia patients from healthy controls, significantly outperforming various traditional machine learning baselines and existing comparative models. These findings validate the effectiveness of encoding topological structures into images combined with color-equivariant modeling strategies, offering a novel perspective for intelligent identification of neuropsychiatric disorders.
Introductory programming is particularly challenging for students from non-computer majors, not only because of syntactic complexity but also due to the lack of a cognitive transition from everyday reasoning to algorithmic problem solving. This study presents a series of instructional episodes from an introductory C programming course to illustrate how abstract programming concepts can be grounded in familiar real-life experiences. These episodes show how students transitioned from intuitive understanding to structured reasoning and ultimately to independent program construction. Classroom observations indicate increased willingness to initiate coding, stronger persistence in debugging, and an emerging ability to transfer solution patterns to new problems. The findings suggest that connecting algorithmic concepts with everyday cognitive experiences reshapes students' perception of programming from knowledge memorisation to meaningful problem solving. This narrative provides practical and transferable teaching insights for programming education in similar contexts.
This study systematically evaluates the robustness of the Speech-Paraformer-Large automatic speech recognition (ASR) model under simulated industrial noise and proposes an effective post-processing enhancement strategy for safety-critical voice-controlled human-robot interaction in manufacturing environments. the controlled experiment used a dataset of 10 Mandarin Chinese industrial commands recorded in clean conditions (16 kHz, 16-bit PCM). Noisy test conditions were generated by mixing clean recordings with continuous white noise and authentic industrial machinery noise at Signal-to-Noise Ratios (SNR) from 20 dB to -10 dB (5 dB increments). The pre-trained Speech-Paraformer-Large model was evaluated, and a text-based verification layer with three hierarchical matching strategies (fuzzy exact matching, substring containment, sliding window similarity) was implemented as post-processing; performance was assessed via Word Error Rate (WER) and accuracy across 50 test utterances per condition. Results show that industrial machinery noise is significantly more detrimental to ASR performance than white noise (24% vs. 70% accuracy at -10 dB SNR). The proposed verification layer consistently improved performance across all SNR levels: accuracy increased by 8% (88% to 96%) under 0 dB white noise and by 10 percentage points (24% to 34%, 41.6% relative improvement) under -10 dB industrial noise. It also reduced substitution errors by 34%, insertion errors by 31%, and total errors by 32%, with unexpected efficiency gains (51.2% reduction in computation time at 0 dB industrial noise). This study demonstrates that intelligent post-processing can achieve practical, deployable robustness gains without model retraining or acoustic preprocessing, and the proposed text-based verification layer provides a cost-effective solution to improve voice control reliability in industrial environments, with direct implications for manufacturing safety and efficiency.
We present a data-driven method for constructing a self-evaluating questionnaire that assesses knowledge in usability, user experience (UX), and accessibility. Traditional diagnostic tools in these domains are often long and cognitively demanding, reducing response rates and practical utility. Our approach leverages supervised machine learning methods such as Multiple Linear Regression, Random Forest, XGBoost, and Univariate Feature Selection to quantify the informational value of each question based on its predictability from others. Using this technique, we generate weighted scores that reflect a respondent’s relative expertise and enable real-time ranking among peers. Applied to 153 responses collected from graduate students and professionals, our system demonstrated that the questionnaire could be reduced by up to 82% from 62 to just 10 questions—while maintaining high accuracy in final scores and ranks. This work contributes a scalable, interpretable framework for knowledge assessment in HCI education and practice and supporting efficient evaluation.
Accurate segmentation of Langerhans cell morphological characteristics is of great significance for the diagnosis of corneal diseases and the assessment of activation degree. However, the precise segmentation of corneal Langerhans cells remains unexplored. Manual dense annotation of Langerhans cells is a time-consuming and labor-intensive task. In order to achieve automated and accurate segmentation of Langerhans cells, this paper proposes a prior denoising framework, through boundary-aware refinement module which consists of a random mask dilation strategy to force the model to locate the target by understanding the target features rather than relying on the environment and a shrinking values generation strategy to gradually capture the target contour and improve the robustness of the refinement model. Experiments on one dataset and twelve types of neural networks have proven that our method significantly improves the accuracy of Langerhans cell segmentation and has strong versatility.
This paper explores potential reinforcements of the RSA algorithm and examines several alternative public-key cryptosystems. Building on the mathematical foundation of integer factorization, RSA has been the cornerstone of secure digital communication, yet it faces vulnerabilities from parameter weaknesses and emerging quantum algorithms. To address efficiency and resilience, this study first analyzes multi-prime RSA, highlighting its advantages in decryption speed through the Chinese Remainder Theorem, while also noting its reduced security margins. In addition, the Goldwasser–Micali cryptosystem is evaluated for its probabilistic encryption mechanism, which enhances semantic security by producing randomized ciphertexts. The LUC encryption scheme, based on Lucas sequences, is then discussed as a variant of RSA with potentially stronger resistance against certain attacks. Finally, an algebraic encryption method utilizing polynomial roots is introduced as an innovative approach, though its practical security remains uncertain. Collectively, these explorations illustrate the trade-offs between efficiency, ciphertext size, and security, and point toward future directions in strengthening public-key cryptography against advancing computational threats.
Breast cancer is the most common malignant tumor among women, and histopathological images play a crucial role in the differential diagnosis of benign and malignant lesions. Although Convolutional Neural Networks (CNNs) and Transformers have been widely used in medical image classification, CNNs often struggle to capture global structural patterns, while Transformers may underperform in modeling fine-grained local features. To address this, we propose a Hybrid Gated CNN-Transformer (HGCTransformer) model for breast tumor histopathological image classification. The model introduces a Dual-Branch Convolution and Attention Residual Module (DBCARM) into the Transformer block to integrate local texture and global contextual information. Additionally, a Gated Multi-Scale Feed-forward Network (GMSFN) is designed to enhance the discrimination of multi-scale malignant features, such as nuclear atypia and architectural disarray. Experimental results on public breast histopathology dataset indicate that the proposed method achieves a classification accuracy of 99.76%, underscoring its potential for computer-aided diagnosis of breast cancer.
Foundations of Information Security Mathematics supports subsequent cryptography courses, yet teaching abstract algebra as an isolated theory often prevents students from relating algebraic structures to cryptographic mechanisms. This paper interprets algebra as the computational spaces of cryptography and reorganizes the course through a requirement-driven structure, where cyclic groups, finite fields, and quotient polynomial rings are introduced as progressively extended environments. A computational-space-oriented pathway is implemented by introducing concepts through cryptographic operations and reinterpreting prior discrete mathematics knowledge. Classroom practice shows a shift from procedural to structural understanding and a unified view of different cryptographic schemes, turning the course from a set of prerequisites into the structural foundation for later study.
