University courses with hundreds of students have become common, particularly during early years of university studies. The sheer scale of these courses limits traditional instruction, shifting it towards a one-to-many mode of delivery. This shift reduces student–instructor interaction and tailored instructor feedback which are crucial for student success. Automated feedback systems allow scaling feedback, but they often reduce instructor contributions to student learning. This paper investigates how emerging technologies can support, rather than replace, instructors in tailoring their teaching and feedback to identify and correct student knowledge gaps at scale. To address this challenge, the paper introduces a novel technological solution: the Knowledge Gaps to Mastery (KG2M) approach. KG2M combines discussion forum data with course-specific content and leverages large language models (LLMs) and Retrieval-Augmented Generation (RAG) for the dual purpose of identifying prevalent class-level knowledge gaps and transforming them into targeted learning activities and formative assessments. The approach was deployed across three computer science courses with a combined enrollment of 1,355 students and evaluated through semi-structured interviews with five instructors. Results indicate that instructors found the tool intuitive and pedagogically valuable, particularly for surfacing knowledge gaps and generating actionable teaching insights. The paper reports on the tool, the evaluation, and the current limitations of the approach that emerged during instructor evaluation. • Large classrooms are prevalent at universities, but they hinder instructors’ ability to provide feedback tailored to students’ needs. • The rise of GenAI shows promise in providing automated feedback at scale; however, it often lacks the nuanced guidance of instructor-led feedback. • Leveraging discussion forums with appropriate design and tech innovation could minimize instructor workload and address students’ learning needs. • We introduce an approach (KG2M) using LLMs and RAG to spot class-wide knowledge gaps and generate learning activities. • Evaluation via case studies in 3 CS courses (1355 students and 2878 unique posts) and with 5 instructors emphasizes pedagogical value.

Scientific Foundation Models (SciFMs) represent a transformative paradigm for addressing complex scientific and engineering problems by leveraging large-scale pretraining and deep learning architectures. Unlike traditional numerical solvers, which require problem-specific discretization and extensive parameter tuning, SciFMs aim to learn generalizable representations of physical laws, thereby enabling broad applicability with minimal retraining. In the absence of a rigorous definition, this work categorizes their capabilities into four key dimensions-domain adaptation, domain generalization, problem adaptation, and problem generalization; thereby providing rigorous definitions for SciFMs, which we refer to as the sufficient conditions. To operationalize these conditions in practice, we further introduce the necessary conditions. We also propose a taxonomy of SciFMs based on model architecture, learning objectives, and training strategies. Although SciFMs demonstrate remarkable potential in computational science and engineering, several challenges persist. Ensuring physical consistency, interpretability, and robustness under extrapolation to extreme regimes remains a critical hurdle. Furthermore, the significant computational demands and the lack of standardized benchmarks present substantial barriers to widespread adoption. This work surveys existing SciFMs across diverse domains-including chemistry, materials science, biology, climate and weather, Earth observation, geophysics, chaotic dynamics, robotics and control, nuclear science, etc., while categorizing them under the proposed definitions. We further outline open research directions, emphasizing the integration of domain knowledge with data- and physics-driven approaches and the development of efficient architectures to enhance generalization. Addressing these challenges will be critical for SciFMs to accelerate scientific discovery and deliver real-world impact. The GitHub repository for this work is available at https://github.com/ParamIntelligence/Awesome-Scientific-Foundation-Models.

• Multi-sector investigation into data quality dimensions and impacting factors. • A new theoretical framework to support organisational digitalisation or diversification. • A three-step process to apply the framework is presented to evaluate data quality and identify areas for improvement. • Thematic analysis of interviews with 31 practitioners from five sectors. • Four segments of data quality dimensions and five impacting factors are identified. The quality of data is central to decision making across all sectors. However, the many single-application studies, proposing hundreds of dimensions, make data quality assurance a daunting prospect. The diversification of industries is compounding this challenge, making a multi-sector classification essential. We propose a new classification for data quality – the DaTUM framework – which can act as a starting point for data quality assurance across diverse sectors. The framework was developed from reflective thematic analysis of 31 in-depth semi-structured interviews with academics and industry professionals from engineering, policy, economics, computer science, and psychology domains who generate, process, or analyse data. Eleven data quality dimensions and five factors that impact data quality were identified. The DaTUM framework, along with the operationalisation process, will simplify digitalisation and diversification efforts within organisations and support targeted data management and quality improvement strategies which are vital to achieving the true value of data.

Unpacking ethics-domain of intelligent-TPACK scale in relation to in-service teachers’ trust and distrust

Integrating artificial intelligence with experimental and computational materials science to predict crystallization in multicomponent glasses

Artificial Intelligence (AI) is changing and revolutionizing today’s economy and work life. A basic understanding of this technology is beneficial for even non-technical roles, highlighting the interdisciplinary nature of AI and its diverse applications. However, it is difficult to create significant, practically relevant learning experiences related to AI for students of different backgrounds, especially for students outside of computer science programs. To tackle this problem, we designed and evaluated an interdisciplinary, project-based course combined with creativity methods, where students from diverse study programs worked on an everyday challenge and tried to build AI prototypes to address it. The interdisciplinary nature of the course enabled students from diverse disciplines to collaborate and learn from one another. The course focused on project-based learning, providing students with hands-on experience in AI product design and implementation. It also incorporated teamwork and collaboration activities that enabled students to gain a better understanding of AI jointly. The course has been conducted and evaluated across two consecutive editions, involving a total of 32 students, providing a robust basis for the analysis presented in this study. Overall, the student feedback was favorable, indicating an enhanced sense of confidence in their AI abilities. We provide evidence that a project-based, interdisciplinary AI course incorporating creative methods can be an effective approach for students from diverse academic backgrounds to expand their knowledge and gain a more nuanced grasp of AI technologies. • A reusable university course design to teach AI and programming competencies • Encouraging students to apply creativity and innovation methods during project work • Fostering student collaboration, communication, and effective teamwork • Lessons learned and takeaways to support other AI educators

This systematic review synthesizes 64 empirical studies to examine how Generative AI (GenAI) shapes learning in Computer Science Education (CSE), particularly in programming, debugging, algorithmic reasoning, and computational problem-solving contexts. Grounded in Constructivist, Sociocultural, Cognitive Load, Adaptive Learning, and Metacognitive Learning theories, the review adopts an integrative perspective to analyze how GenAI-driven adaptivity, AI output qualities, hallucination dynamics, and cognitive–affective regulation influence learners’ interpretation, cognitive processing, and learning outcomes. Findings reveal a dual impact of GenAI in CSE. On the negative side, hallucinated or misleading outputs can increase extraneous cognitive load during programming and debugging and promote over-reliance on system-generated content. They may also perpetuate inequities due to limited access in low-resource settings or insufficient support for culturally and linguistically diverse learners. These effects can disrupt error detection, self-monitoring, and problem-solving, leading to impaired learning performance, and widened educational disparities. On the positive side, when embedded within structured, equitable, and pedagogically grounded environments, GenAI supports reflective programming practice by promoting self-monitoring, verification, and strategic adjustment, thereby enhancing problem-solving skills, engagement, and personalized learning outcomes. By framing learning performance, hallucination dynamics, and problem-solving as interconnected dimensions of GenAI-supported computing education, this review provides a theoretically coherent and pedagogically grounded lens for understanding how GenAI reshapes learning in CSE. The review’s novelty lies in its integrative conceptual framework, offering actionable insights for designing equitable, cognitively balanced, and instructionally effective GenAI-supported learning environments.

Supporting K-5 computer science integration through high-quality teacher professional development

The situated nature of a computer science teacher's persistent engagement over a decade: A lines of practice perspective of learning, teaching, and leading

What undergraduate students need to know and actually know about generative AI

Ecological impacts of hydropower projects in the lower reaches of the Yarlung Zangbo River driven by machine learning: Focusing on nitrogen and phosphorus cycle prediction and assessment

Nonlinear equations frequently appear in diverse fields of applied sciences, where real-world phenomena cannot be accurately represented by linear models. Therefore, developing efficient numerical methods to approximate the roots of such equations remain a challenging and intellectually stimulating task. These methods are crucial in physics, engineering and computer science for solving nonlinear equations. In response to the growing demands of real-time systems, complicated simulations and high-performance computing, this article introduces few novel root-finding methods that significantly improve the convergence order of the traditional approaches. Accelerated decomposition technique is to diversify different classes of iterative methods. Newly derived methods are compared with existing methods numerically as well as graphically. Polynomiography is employed to visualize the basins of attraction, providing insight into the convergence behavior and stability of the methods. The results indicate that the new algorithms not only overcome the limitations of existing techniques but also offer a visually intuitive understanding of root-finding processes. This study presents innovative root-finding methods that utilize accelerated decomposition techniques. The proposed methods demonstrate a significant improvement in convergence order compared to traditional approaches Through numerical and graphical comparisons, the newly derived methods are shown to outperform existing methods. .

The vascular system serves as the central architecture of the human circulatory network, whose structural and functional integrity is vital for maintaining homeostasis and is closely associated with the development and progression of major diseases, including cardiovascular diseases, cerebrovascular diseases, hepatic disorders, ocular diseases, and renal conditions. Traditional single-dimensional research models present evident limitations in deciphering the complex vascular system. In this context, this article formally introduces the concept of "Vasomics". Vasomics is an emerging omics discipline that integrates clinical medicine, basic medicine, biology, computer science and artificial intelligence to systematically analyze the vascular system using multimodal and cross-scale approaches. This article elaborates on Vasomics from six key aspects: its background, core technologies and phenotyping, methodological framework, research progress, applications, and challenges and prospects. By enabling the integration of multi-scale vascular phenotypes from macroscopic to microscopic levels, Vasomics is poised to offer a new paradigm for deciphering vascular health and disease.

Innovative products that receive favorable reviews but never catch on with consumers belong to a category known as the "best game no one played." We combined an adoption threshold model with an opinion dynamics model to examine reasons why certain high-quality products and ideas never achieve expected levels of commercial success. Computational social scientists use opinion dynamics models to analyze consensus formation, and adoption threshold models to study acceptance scenarios. However, most studies based on the first type focus on opinion exchanges without discussing follow-up actions, and most based on the second type only examine ways that individual decisions are dependent on numbers or proportions of friends and neighbors already engaged in specific behaviors, regardless of opinion differences. For this study, four kinds of theoretical networks (regular lattice, random, small-world, scale-free) served as underlying social network structures, and an agent-based simulation approach was used to analyze opinion exchange dynamics and product acceptance. Results indicate that computational agents were capable of changing pro/con opinions regarding issues, products, policies, etc. based on communication with neighboring agents via underlying social networks, and of making acceptance/rejection choices based on a combination of individual adoption threshold plus observations of their neighbors' behaviors. A series of sensitivity analysis simulation experiments was conducted to identify model-related factors, determine non-linear correlations among them, and quantify degrees of influence. Factors exerting the strongest influence or requiring greater care when applied to cases of innovation diffusion were examined. Sensitivity analysis results indicate that agent adoption threshold mean exerted the greatest influence, followed by agent attitude mean and bounded confidence. Mechanism decomposition experiments revealed that the testimony effect neutralizes opinion clustering, making coordination failure the dominant driver of the opinion-adoption gap. These findings yield predictions distinguishing the model from information cascades, network externalities, and global games.

ABSTRACT Background and Context Digital technologies have become part of our lives. Meanwhile, computer programming, the activity that mediates the communication between humans and technology, has expanded to fields and practitioners beyond Computer Science. Programmers who give instructions to digital artifacts should have proper awareness of the artifacts’ states and environments. This often requires them to change their point of view in order to better interact with those systems. Notably, this skill resembles a fundamental human ability that can be learned and developed over time: empathy. Objective In this study, we probed potential connections between computer programming activities (object-oriented programming, educational coding games, educational robotics) and programmers’ level of empathy, to gain further insights into human-technology interactions. Method Through a quantitative approach, an online survey was implemented and data were collected from 104 adults in Greece with prior experience in computer programming. The majority of the participants work in education or education-related fields (58%). Findings According to the results, statistically valid relationships were observed between ease of programming and empathy, either directly with it or with at least one of its components (cognitive empathy, emotional empathy and social skills). Implications The indicated correlations between computer programming and empathy could represent a paradigm shift in the field of Education and lead to a deeper understanding of Computer Science Education and Didactics.

ABSTRACT Background and Context Despite increasing momentum to integrate computer science (CS) into K-12 education, elementary teachers often report low confidence in teaching CS due to limited experience, resources, and pedagogical knowledge. Teaching self-efficacy, which refers to the belief in one’s ability to teach effectively, plays a critical role in teachers’ willingness to adopt CS instruction. There remains limited research on how professional development (PD) programs improve elementary teachers’ CS teaching self-efficacy, particularly regarding the specific sources that contribute to their confidence. Objective This study aimed to investigate the impact of an integrated CS PD program on in-service elementary teachers’ CS teaching self-efficacy. It also sought to identify which aspects of the PD and learning experiences served as key sources of efficacy information and how teachers perceived their relative importance. Method Using a mixed-methods approach, we collected data from 33 in-service elementary teachers. Quantitative data were gathered through pre-post administration of the CSTEBI, analyzed using paired-samples t-tests. Qualitative data from teacher reflections were coded by two researchers to identify themes related to sources of teaching self-efficacy. Findings The PD significantly improved teachers’ CS teaching self-efficacy. Qualitative analysis identified simulated modeling as the most influential source of efficacy, followed by cognitive pedagogical mastery, cognitive content mastery, cognitive self-modeling, and effective actual modeling. We also identified the integration of disciplines as a novel and distinct source of efficacy. Implications Our findings suggest that effective CS PD should prioritize experiential learning approaches where teachers engage as learners while simultaneously developing both content knowledge and pedagogical content knowledge, particularly through cross-disciplinary applications.

Large public buildings are characterized by high occupancy and complex functions, which can easily lead to issues such as congestion in evacuation routes and disorderly crowd behavior in the event of a fire. Conducting analyses and simulation assessments of evacuation mechanisms in fire scenarios is of great significance for improving building fire safety standards and emergency management capabilities. At present, there are significant gaps in the existing research on evacuation simulation for public buildings: Most studies rely on fixed parameter assumptions and fail to effectively quantify the influence of subjective factors such as psychological factors, safety awareness, and social roles on evacuation behavior. Moreover, the combination of BIM technology and evacuation simulation mostly focuses on the presentation of spatial geometric information, lacking a deep integration with quantitative methods for quantifying the subjective behavior of personnel, resulting in insufficient authenticity and predictive reliability of evacuation simulations, and making it difficult to precisely support fire protection design and emergency decision-making. In response to this research gap, this study has established an integrated framework that quantifies subjective human-related factors, maps them to key behavioral parameters through fuzzy inference, and couples them with BIM-based fire and evacuation simulations to provide a verifiable linkage between fire scene constraints, human behavior, and evacuation outcomes. This paper employs fuzzy logic theory together with Pyrosim and Pathfinder to investigate the effects of human-related subjective factors and fire scene conditions on fire evacuation safety. A questionnaire survey was conducted to examine how psychological factors, safety awareness, and social roles of pedestrians influence evacuation behavior. Through the reliability and validity test of the valid questionnaire data and the spearman correlation analysis, it is found that there is a significant positive correlation between safety awareness, psychological factors, social roles and the evacuation behavior. Based on fuzzy rules, the domains and membership functions of the linguistic variables representing these factors are defined, enabling the quantification of the influencing factors and the calculation of the initial evacuation speed. Finally, a BIM model was established and applied to the evacuation simulation of a large shopping mall project in Southwest China to verify the feasibility of the fuzzy algorithm and the safety of the evacuation design. This research innovatively combines fuzzy algorithms with BIM technology, making up for the deficiencies of existing studies in the quantification of subjective factors of personnel and the deep integration of BIM technology. It provides a more scientific calculation plan and data for the study of public building evacuation, and offers reference basis for fire protection design, personnel allocation, emergency plan formulation, and rescue operations.

ABSTRACT Background and Context Students in Computer Science (CS) tend to have high dropout rates. The variables related to dropout have been studied, but no consensus has been found regarding the most important influence factors including university admission requirements that can be variable for different countries and universities. Objective This empirical study examines the variables which predict whether CS students graduate within the nominal study period, and whether variables in models differ for female and male students. Method Data of 390 CS students’ personal characteristics, academic preparation before the university and academic achievement during the first year were included in this quantitative study as potential variables predicting completion of CS studies within the nominal study period. The data were analysed using binary logistic regression. Findings Graduation within the nominal study period was influenced by factors such as student age, gender, higher secondary education graduation with honours, and the percentage of curriculum completion in the second semester. For both female and male students, higher secondary education graduation with honours and the percentage of curriculum completion in the second semester emerged as the key predictors of timely graduation. Implications The CS admissions processes should reconsider placing greater weight on honours graduation. Second-semester academic progress should be used to trigger early-alert systems and offer proactive academic support. Interventions should be customized to offer flexible schedules and personalized advisory options for older students and provide additional monitoring for male students.

Learning about the causal structure of the world is a fundamental problem for human cognition. Causal models and especially causal learning have proved to be difficult for large pretrained models using standard techniques of deep learning. In contrast, cognitive scientists have applied advances in our formal understanding of causation in computer science, particularly within the causal Bayes net formalism, to understand human causal learning. In the very different tradition of reinforcement learning (RL), researchers have described an intrinsic reward signal called 'empowerment' which maximizes mutual information between actions and their outcomes. Empowerment may be an important bridge between classical Bayesian causal learning and RL and may help to characterize causal learning in humans and enable it in machines. If an agent learns an accurate causal world model, they will necessarily increase their empowerment, and increasing empowerment will lead to a more accurate causal world model. Empowerment may also explain distinctive features of children's causal learning, as well as providing a more tractable computational account of how that learning is possible. In an empirical study, we systematically test how children and adults use cues to empowerment to infer causal relations and design effective causal interventions. This article is part of the theme issue 'World models in natural and artificial intelligence'.

In recent years, artificial intelligence (AI) has increasingly influenced daily life and scientific research. Traditionally, AI-related courses have targeted computer science majors, while systematic instructional opportunities for early-stage undergraduates from non-computing backgrounds remain limited. To bridge this gap, we developed an AI course that integrates project-based learning with large language models (LLMs). Specifically, we designed four progressive assignments based on our research project (i.e., drug–drug interaction network clustering analysis). The course does not require prior knowledge of pharmacology or programming. Instead, LLMs are used as assistive tools to support programming, data analysis, and result interpretation. Students engage in a complete workflow, including data curation, algorithm implementation, and critical evaluation of results. Preliminary feedback shows that this approach supports the development of problem-solving skills and increases student engagement. This study provides a framework for integrating LLMs into project-based learning. We believe that this teaching proposal will be valuable and inspiring for educators seeking to design or enrich similar courses.