In recent years, quadruped robots have received many attention due to excellent adaptability in complex terrains,and the key to their stable locomotion lies in gait coordination. But,the traditional central pattern generator (CPG) models often face challenges such as high reliance on manual experience for tuning coupling parameters and poor adaptive capability. To address this problem, this study proposes a control method integrating coupling dynamics modeling and intelligent optimization. And,a four-leg coupling dynamics model based on Hopf nonlinear oscillator is constructed, in which coupling matrix describes inter-leg phase relationships. The matrix is automatically optimized by incorporating a genetic algorithm and implementing global search with a phase synchronization stability metric as the fitness function. Simulation results show that the optimized coupling parameters significantly improve the phase coordination ability of the four-leg oscillators. This effective eliminates phase deviations under natural dynamics, and greatly enhances both gait synchrony and stability.And so,the study contributes to the autonomous adaptability of quadruped robots by providing a data-driven global optimization framework for their gait control.\

In the digital economy, traditional software engineering faces complex needs and fast change, and it often meets risks such as cost overruns and low process efficiency. AI methods, including machine learning and natural language processing, provide strong support for a more intelligent form of development. This paper reviews the path of AI use in software work through literature study and real cases, and it covers tasks from requirement analysis to testing and operation. The findings show a shift from a simple helper to a driver of the whole process, and cases from many fields report higher efficiency and better quality. Yet several problems remain, such as weak model transparency, unstable data, and a lack of skilled staff. It presents a framework and guides firms toward high-quality growth globally.

This study takes Duolingo as a case study and, drawing on Second Language Acquisition (SLA) theories, examines the role of language-learning applications in the construction of a systematic language learning framework. It systematically analyzes the impact of interface interaction design on second language learning processes. Through an examination of Duolingo’s User Interface Design principles, learning task organization, and interactive feedback mechanisms, this study summarizes the app’s strengths in facilitating language input, enhancing learners’ attentional focus, and sustaining learning motivation. The study further identifies several limitations of Duolingo, including insufficient systematic representation for less commonly taught languages, limited support for productive skills, and a lack of coherence in learning pathways. Overall, this research provides theoretical insights for interface design and curriculum development in language-learning applications.

Medical time-series data prediction requires exact methods that enable doctors to track patient health and detect medical issues early. The heart’s electrical signals, which the electrocardiogram (ECG) records, show intricate time-based patterns because they contain nonlinear elements and include brief intense changes. The ECG time-series modeling process often use LSTM networks because these networks effectively track extended temporal patterns which exist in the signal data. The performance of LSTM models heavily depends on hyperparameter selection, and researchers who manually choose parameters often end up with oversmoothed predictions which fail to include vital clinical waveform characteristics. This research proposes Particle Swarm Optimization–enhanced LSTM (PSO-LSTM) framework as a new solution for ECG time-series prediction. The PSO algorithm conducts automatic hyperparameter optimization through its process of optimizing hidden layer sizes and network depth and learning rate values while maintaining the core LSTM structure. The MIT-BIH Arrhythmia Database provides ECG signals for experiments which undergo identical preprocessing operations and evaluation procedures throughout all testing runs. The proposed PSO-LSTM model is compared with a baseline LSTM under identical training conditions. The research evaluates prediction accuracy by using three performance metrics which consist of mean squared error (MSE) and root mean squared error (RMSE) and mean absolute error (MAE). Results show that the PSO-LSTM model achieves lower prediction errors across all metrics compared to the baseline LSTM. The research shows that metaheuristic-based hyperparameter optimization techniques boost LSTM prediction accuracy for medical time-series data through optimization methods which do not require additional model components.

Traditional surgical training is confronted with inherent bottlenecks including inadequate cadaver supply, exorbitant costs, and subjective performance evaluation. Although virtual reality (VR) technology enables the provision of an immersive training milieu, its fidelity, realism, and intelligent capabilities remain constrained. The integration of artificial intelligence (AI) is profoundly revolutionizing multiple dimensions of VR-based surgical simulation. This study systematically summarizes the core technologies and practical applications of AI in VR-assisted surgical training and simulation. We propose a classification framework based on technical modules, delineating the roles of AI into three core dimensions: (1) AI-driven virtual surgical scene generation; (2) AI-enhanced physics engine to achieve real-time and realistic tissue deformation and interaction simulation; and(3) AI-enabled skill assessment and feedback. This study further elaborates on the predominant challenges faced by current technologies, such as physical-visual inconsistency, small sample data issues, and insufficient clinical validation, while outlining prospective research orientations such as multimodal sensing, federated learning, and AI-empowered robotic surgery skill transfer.

Human–Computer Interaction (HCI) is an interdisciplinary field that integrates insights from computer science, psychology, cognitive science, and design studies. As digital technologies such as artificial intelligence, mobile computing, and multimodal sensing continue to advance, human–technology interaction has become increasingly complex, extending beyond traditional concerns of usability toward cognitive, affective, and social dimensions. This study presents a structured literature review of HCI, aiming to clarify its conceptual development, theoretical foundations, and practical applications. The review traces the evolution of HCI from early command-line and graphical interfaces to mobile, context-aware, and intelligent multimodal interaction paradigms. It further examines key theoretical and computational frameworks, including User-Centered Design, context-aware interaction models, and multimodal fusion mechanisms, highlighting their roles in supporting adaptive and human-centered interaction. In addition, representative applications in healthcare, transportation and autonomous driving, and service and humanoid robotics are analyzed to illustrate how these frameworks are instantiated under domain-specific constraints. Overall, the findings indicate a clear shift in HCI toward adaptive, personalized, and continuous interaction, emphasizing collaboration between humans and intelligent systems.

Multimodal understanding, which requires models to jointly reason over visual and linguistic information, has become a core challenge in artificial intelligence (AI). Visual Question Answering (VQA) stands as a paradigmatic task for investigating these multimodal reasoning capabilities. While early VQA systems relied on task-specific architectures, recent breakthroughs in Multimodal Large Language Models (MLLMs) have significantly reshaped the field by proposing unified, instruction-driven multimodal reasoning frameworks. By conducting a systematic literature review, this paper scrutinizes the evolution of VQA from traditional CNN–LSTM-based models to modern MLLM-based approaches. The review centers on representative architectures and training paradigms, including BLIP-2 and LLaVA, to analyze how large language models and pretrained vision encoders are integrated for flexible and open-ended visual reasoning. In addition, this paper identifies and deliberates on critical challenges confronting contemporary MLLMs, encompassing modality imbalance, insufficient cross-modal alignment, and hallucinations. This paper concludes that while MLLMs have substantially expanded the application scope and functional capabilities of VQA systems, they still grapple with reliable visual grounding and balanced multimodal fusion. Addressing these limitations is paramount for constructing trustworthy and robust VQA systems, and future research should prioritize improving alignment mechanisms and mitigating hallucinations in multimodal reasoning.

Online gaming industry has expanded rather rapidly in the last 10 years, which has intensified competition between the developers and the single games have become harder to distinguish on the market. Although a substantial amount of current literature analyzes the behavior of players, including their persistence and attendance, very little research is devoted to the features of the very games. The given study is devoted to game titles and examines the aspects of their design and features concerning the market in reference to the degree of popularity. Based on the data gathered on 27,075 games on the Steam platform, each one of the titles is classified as Low, Medium, or High popularity using a composite engagement measure. The data is explored and converted into information about the pricing, content area, achievement and activity of players before creating the model. Random Forest model is then used and reaches an accuracy of 66.24 which is greater in comparison to the one achieved using the logistic regression baseline. Its achievement is more accurate on the high-popularity category games where F1-score is 0.780. The achievement-related variables and pricing measures, game age, content scope are likely to be near the top in the feature importance rankings. This tendency suggests that popularity is not a characteristic that is linked to any one defining factor, instead, it is a complex of factors that are combined.

As civil aviation evolves toward Single Pilot Operations (SPO), the situational awareness of intelligent pilot assistance systems is critical for flight safety. Specifically, a precise, real-time understanding of pilot behavior is indispensable for compensating for the absence of "human redundancy" in the cockpit. Pilot behaviors exhibit significant spatiotemporal correlations, encompassing instantaneous operational actions (e.g., control stick manipulation) and continuous physiological poses (e.g., fatigue and observation). However, existing research often isolates these into independent single tasks, neglecting the intrinsic coupling mechanism between them in biomechanical and spatiotemporal dimensions. This results in computational redundancy, violating the stringent real-time and low-power constraints of airborne embedded platforms. To bridge this gap, this paper presents a review of concurrent recognition techniques for cockpit behaviors based on Multi-Task Learning (MTL). First, a hierarchical taxonomy of cockpit behaviors is constructed, analyzing the mechanisms underlying the correlation between actions and poses. Second, the evolution of concurrent recognition architectures is discussed in depth, with a focus on the shared feature-extraction backbone that marks the shift from 3D Convolutional Neural Networks (3D CNNs) to Spatiotemporal Transformers. Furthermore, joint optimization strategies for heterogeneous tasks, such as hard parameter sharing mechanisms and homoscedastic uncertainty weighting, are analyzed. Finally, datasets and cross-domain transfer methods are summarized. Future challenges, including environmental robustness, lightweight model deployment, and interpretability, are discussed to provide theoretical references and technical support for the construction of next-generation human-machine collaborative monitoring systems.

Large Language Models (LLMs), such as GPT, have recently been deployed as agents in interactive environments, including games. While these models exhibit strong natural language understanding capabilities, they often produce incorrect or inconsistent decisions in strategic games that require strict rule adherence and multi-step planning. This project investigates the decision-making behavior of an LLM-based agent in the game of Tic-Tac-Toe. A simulator was developed to enable gameplay between the LLM and a human player, and multiple prompt designs were evaluated. The experiments compare prompts with no contextual guidance, basic rule descriptions, example-based contexts, and explicit strategic instructions. The results demonstrate that prompt design plays a critical role in improving the reliability of LLM-based game agents and offer insights into how decision-making accuracy can be significantly enhanced without modifying the underlying model.