Source code plagiarism detection plays an indispensable role in computer education. Over the past few decades, various tools and systems have been developed to solve software plagiarism detection with relatively satisfactory results. However, hardware description language (HDL) plagiarism detection has become a new challenge with the popularity of FPGA-based digital circuit design courses. Constructing a Verilog HDL plagiarism detection approach that supports fast incremental detection on large samples has become a thorny issue. This paper proposes a scalable approach to achieve Verilog HDL plagiarism detection that supports incremental detection on large samples. The method employs a dual-layer detection strategy at both the submission-level and the module-level. At the submission-level, it conducts project-wide comparisons, while at the module-level, it calculates a project's suspicious score through a prefix matching-based risk assessment mechanism. This approach effectively distinguishes common modules from potential plagiarism, significantly reducing the false positive rate. Our method detects more plagiarism pairs with higher accuracy than the widely used Moss (Measure of Software Similarity) system. To further validate the effectiveness of our approach, we conducted a comparative analysis with an ANTLR-based syntax-aware detector. On the dataset of Verilog HDL projects from a computer organization course (Fall 2023 and Fall 2024), our method achieved higher precision and reduced time consumption compared to the ANTLR-based model. Initially, this study leverages filtered token features and the locality-sensitive hashing algorithm to transform Verilog HDL projects into fingerprints. An efficient index is constructed to accelerate both incremental detection and batch detection. Four strategies for filtering N-Gram token sequences are used in this study. The proposed fast incremental Verilog HDL plagiarism detection approach has a logarithmic level of time complexity. Experiments were conducted to find and validate the optimal parameters for the proposed strategies on real-world datasets. The experimental results show that combining our method with Moss increases the convenience and accuracy of Verilog HDL plagiarism detection in real courses.

Blended learning enriches students' experiences across diverse environments while generating multimodal data related to learning activities. However, it presents challenges in the appropriate use of multimodal data to track students' performance development. Previous models with fixed-length inputs or static fusion mechanisms inadequately model temporal dependencies across behavioral modalities. In this article, we integrate variable-length time series over weeks to forecast performance for the subsequent week. As the main contribution, we propose a two-stage training model that relies on a transformer for temporal attention-based multimodal fusion. We conducted experiments on two real-world datasets, FC2023 and CS2023, derived from hybrid mode courses involving 439 and 199 students, respectively. The results demonstrate that multimodal fusion yields better periodical prediction compared to the unimodal approach. Ultimately, aiming at predicting the week-by-week development of student performance, the proposed model achieves the area under the curve of receiver operating characteristic of 81.02% on FC2023 and 82.65% on CS2023. This approach, which leverages multimodal learning analytics, helps educators track each student's learning progress more effectively, enabling the timely implementation of instructional interventions and enhancing educational outcomes.

Graph Neural Networks (GNNs) have been widely adopted due to their strong performance. However, GNN training often relies on expensive, high-performance computing platforms, limiting accessibility for many tasks. Profiling of representative GNN workloads indicates that substantial efficiency gains are possible on resource-constrained devices by fully exploiting available resources. This paper introduces A3GNN, a framework for affordable, adaptive, and automatic GNN training on heterogeneous CPU-GPU platforms. It improves resource usage through locality-aware sampling and fine-grained parallelism scheduling. Moreover, it leverages reinforcement learning to explore the design space and achieve pareto-optimal trade-offs among throughput, memory footprint, and accuracy. Experiments show that A3GNN can bridge the performance gap, allowing seven Nvidia 2080Ti GPUs to outperform two A100 GPUs by up to 1.8X in throughput with minimal accuracy loss.

In blended learning, the learning management system (LMS) records various data related to learning behavior. In most cases, the data extracted from LMS can be used to integrate a learning analytics dashboard (LAD). However, current research focuses more on LAD design and less on the quantitative analysis of the impact of user behavior. To fill this void, the study aims to reveal the correlation between viewing frequency of LAD and learning engagement. In this research, we conducted a learning behavior-based pedagogical intervention for a course in the fall of 2024 and delivered LAD to 325 undergraduates. During the intervention period, the course platform updated the dashboard every week based on the latest learning behavior data, and the data were analyzed based on propensity score matching (PSM). The findings showed that the viewing frequency is positively related to learning engagement. Additionally, the view frequency is independent of the two dashboard interfaces we proposed.

Developing deep learning models on tiny devices (e.g. Microcontroller units, MCUs) has attracted much attention in various embedded IoT applications. However, it is challenging to efficiently design and deploy recent advanced models (e.g. transformers) on tiny devices due to their severe hardware resource constraints. In this work, we propose TinyFormer, a framework specifically designed to develop and deploy resource-efficient transformers on MCUs. TinyFormer mainly consists of SuperNAS, SparseNAS and SparseEngine. Separately, SuperNAS aims to search for an appropriate supernet from a vast search space. SparseNAS evaluates the best sparse single-path model including transformer architecture from the identified supernet. Finally, SparseEngine efficiently deploys the searched sparse models onto MCUs. To the best of our knowledge, SparseEngine is the first deployment framework capable of performing inference of sparse models with transformer on MCUs. Evaluation results on the CIFAR-10 dataset demonstrate that TinyFormer can develop efficient transformers with an accuracy of $96.1\%$ while adhering to hardware constraints of $1$MB storage and $320$KB memory. Additionally, TinyFormer achieves significant speedups in sparse inference, up to $12.2\times$, when compared to the CMSIS-NN library. TinyFormer is believed to bring powerful transformers into TinyML scenarios and greatly expand the scope of deep learning applications.

In programming teaching, teachers or teaching assistants often need to spend a lot of energy helping students solve the problems they face when doing programming. It will be helpful to provide students with valuable programming feedback, such as information on faulty lines. However, most existing algorithms do not perform well on novice programs. Therefore, considering the background in programming teaching, we proposed a novel approach combining static analysis with dynamic detection by using the correct programs submitted by previous students and coverage information for the incorrect program. In particular, the core of the static analysis module is to locate specific faulty lines through syntax tree difference comparison, which includes matching similar programs, variable mapping and replacement, and fault localization based on abstract syntax tree differences. The core of the module on dynamic detection is to perform traditional Spectrum-based Fault Localization. To evaluate the effectiveness of our proposed approach, we conducted some empirical studies on 223 student-failure programs in the real world. The experimental results indicate that our approach outperforms other baselines regarding TOP-1 and TOP-3. Furthermore, we analyzed the performance of our method on different categories of programming problems as well as the effectiveness of the combination of static analysis and dynamic detection.

Automatic program repair (APR) tools are valuable for students to assist them with debugging tasks, since program repair captures the code modification to make a buggy program pass the given test-suite. However, the process of manually generating catalogs of code modifications is intricate and time-consuming. This paper proposes Contextual Error Model Repair (CEMR), an automated program repair tool for introductory programming assignments. CEMR is designed to learn program code modifications from incorrect-correct code pairs automatically. Then, it utilizes these code modifications along with CodeBERT, a generative AI, to repair students' new incorrect programs in the same programming assignment. CEMR builds on the observation that code edits performed by students in pairs of incorrect-correct code can be used as input-output examples for learning code modifications. And the key idea of CEMR is to leverage the wisdom of the crowd: it uses the existing code modifications of incorrect-correct student code pairs to repair the new incorrect student attempts. We chose three of the most related automatic program repair tools, Refazer, Refactory, and AlphaRepair, as the baselines to compare against CEMR. The experimental results demonstrate that, on public and real classroom datasets, CEMR achieves higher repair rates than the baselines. Through further analysis, CEMR has demonstrated promising effectiveness in addressing semantical and logical errors, while its performance in fixing syntactical errors is limited. In terms of time for repairing buggy programs, CEMR costs approximately half as much as AlphaRepair requires. We opine that CEMR can not only be seen as a program repair method that achieves good results with incorrect-correct code pairs but also be further utilized to generate hints to better assist students in learning programming.

The shift to hybrid teaching during the COVID-19 pandemic brought about a real challenge to predict student performance and conduct timely interventions on at-risk students. This study proposed a deep neural network supporting the early prediction of student performance. Bidirectional LSTM, Global Average Pooling, and TIME MASK structure were utilized in the improved GritNet model. Subsequently, this study optimized the hyperparameters with the aid of the hill-climbing algorithm. Finally, on-campus data sets were used in experiments to evaluate the model's performance. Data were collected from a course that carried out multiple iterations from Fall 2021 to Fall 2022. In Fall 2022, the proposed model achieved a ROC-AUC value of 95.47% in the 8th week, while the baseline model only achieved 91.44% in the same week. Besides, the proposed model achieved a ROC-AUC value of 89.67% in the 4th week, which meant it had acceptable prediction performance in the early stage. The experimental findings demonstrated that the model was capable of predicting the academic performance of students in hybrid courses early on.

Small private online courses (SPOCs) have influenced teaching and learning in China's higher education. Learning management systems (LMSs) are important components in SPOCs. They can collect various data related to student behavior and support pedagogical interventions. This research used feature engineering and nearest neighbor smoothing models to predict the performance of students. Five learning behavior features were selected based on Spearman's rank correlation coefficients with students’ final grades. Through testing with data from the fall semester of 2020, the model attained the highest ROC-AUC value of 0.9390. Based on these models, the researchers conducted an engagement intervention that displayed learning behavior dashboards to students in the fall of 2021. During the intervention, the course platform updated the dashboards and notified students weekly. This intervention was further investigated through a randomized controlled trial. The experimental results suggested that the intervention could improve students’ learning behavior in terms of total study time, tutorial reading, and video viewing. In addition, this study used a modified dynamic key-value memory network model to depict a student's knowledge state and to calculate the probability of solving an exercise by mining numerous exercise records. Based on the predicted probability, instructors could recommend personalized exercises for each student. In the fall of 2021, the researchers also conducted a randomized controlled trial on this intervention, demonstrating that this personalized exercise recommendation could increase students’ concept mastery. Experiments revealed that the proposed models and interventions had a positive effect on students’ learning of course content.

Adaptive and intelligent web-based educational systems are made to automate the adaptation of the system to the learners' behaviors and needs. Personalized e-learning platforms should make adaptive adjustments according to the individual students' interactions and their knowledge states (KS). This study proposes a more effective personalized learning path recommendation algorithm to promote the individualized development of students. First, the Dynamic Key-Value Memory Network (DKVMN) is enhanced by integrating a learning behavior module, which is used to trace student knowledge states. Then, the proposed knowledge tracing model is used to simulate virtual students and train recommendation policy based on reinforcement learning (RL). The experimental results show that our personalized learning path recommendation algorithm increases the average knowledge state of students by 12.11% and 5.38% on two different data sets, respectively.

针对学生的学习基础、学习能力、学习效率等差异,从进行课程顶层设计、重构教学内容体系、采取线上线下混合教学、构建多元过程化考核体系4个方面,提出开展大学计算机课程面向赋能教育的个性化教学思路和方法,为学生全面赋能.

This paper presents research that Works in Progress (WIP). Small private online courses (SPOCs) have recently received extensive attention in computing education. In SPOCs, programming exercises are frequently included to train students’ programming skills. Abstract Syntax Tree Retrieval (ASTR) is a system that can help students solve Python problems by inferring the coding goals. However, the coding goal retrieved by ASTR gives students little information about what to do next. In response to this limitation, this work focuses on generating modification hints for students based on the coding goal. In addition, this paper reports on an effort to translate this idea over to Verilog-HDL programming problems. Without any programmed expert knowledge, the final results demonstrate that our system is generally accurate for 1 out of 2 submissions to give hints at a minimum. And for some favorable problems, it potentially performs much better. Furthermore, the results indicate that in the process of retrieval, weighted tree edit distance calculations resulted in improved accuracy over metric tree edit distance calculations.

Due to the COVID-19 pandemic, traditional teaching has been migrated online. Different from traditional face-to-face teaching, when students learn online, the online learning platform will generate various data. And these data make it possible for us to analyze students’ final academic performance. In this paper, we use structural equation modeling (SEM) to analyze the relationship between students’ learning factors. It is found that students’ lab scores (LS), exercise scores (LS) and participation in the discussion (PID) have a direct impact on their final programming scores (FPS). This paper also finds that students’ assignment submissions have an indirect influence on their final programming scores (FPS) but have a direct effect on lab scores (LS). In addition, students’ participation in the discussion (PID) has an indirect influence on assignment submissions (AS) and lab scores (LS). The research in this paper can provide instructional designers with references for instructional design.

This Work-In-Progress paper is in the Innovative Practice category. In the MOOC-related research field, many researchers analyzed students' learning behavior based on the logging data to predict students' performance and improve the course design. Nowadays, Small Private Online Courses (SPOC) are favored in college education, especially in computing education. This hybrid teaching model allows courses to be conducted through Internet, which enables teachers and students to access the course anytime, anywhere. Besides, multimedia resources, including images, videos, and audio could be contained in course materials to strengthen the expressiveness of SPOC. On the other hand, the online learning management system (LMS) collects all the students' interactions with it. But how could we extract meaningful information from them? And how could we improve the learning outcomes of a SPOC? In this study, we analyzed LMS data from a sophomore Computer Structure course. We applied several data mining techniques and conducted an intervention using several visualization techniques. Features were selected according to Spearman's rank correlation coefficient with grades. The correlation coefficient of these selected features ranged from 0.42 to 0.84. Course data were further processed to predict students' performance. The predicted grade was processed in the form of heatmaps to illustrate students' learning behavior. Besides, we further designed an overall view for teachers' perspective, which contains data of all the students in each heatmap. The predicting models were evaluated by ROC-AVC values. Several hyperparameters were tuned in order to pursue better predict performance. The best ROC-AVC value could reach 97.44%.

During the hybrid teaching, knowledge tracing plays an important role in constructing adaptive teaching system. This study models the students' knowledge status by mining a large number of exercise records based on the improved dynamic key-value memory network (DKVMN), which is a knowledge tracing model with two external memory modules. Furthermore, the features of students' behavior are extracted to improve the prediction results of DKVMN. Since the model could depict the evolving knowledge state of students, the visualized results are displayed to both students and teachers. It could encourage students to learn the concepts that have not been mastered. On the other hand, it could help teachers to conduct teaching interventions on the high-risk students.

This Innovative Practice Work-In-Progress presents an integrated signal processing experiment, which can cover most knowledge points of digital signal processing (DSP) course. Since the DSP course focuses on one dimension signal processing, voice signal has a great advantage. We provide an integrated voice signal processing experiment named as Phone Keypad Voice Recognition (PKVR), including the following parts: phone keypad voice collection, Discrete Fourier Transform (DFT) and analysis, filter design, digital query table establishment, number recognition of any keypad voice. Through the improvement of the practice training, the classroom teaching theory can be better understood in an interesting way for our students.

COVID-2019 outbroke in China at the beginning of 2020. Traditional face-to-face teaching cannot be achieved during the epidemic, and all schools across the country turned to online teaching. However, online education still faces many difficulties, and the most important question is how to ensure teaching quality. In this paper, we discussed some pedagogical approaches, including online teaching with the enhanced design of interactions, learning theories and problem-based learning, which are useful for designing courses. Then, a computer foundation course was taken as an example to show how to transit a traditional face-to-face course into an online course utilizing the pedagogical approaches and conduct teaching adjustment based on students' learning behavior.

This Research to Practice WIP presents a UAVs visual navigation based comprehensive experiment to enhance multidisciplinary engineering ability in Aerospace engineering education. In traditional courses, aerospace-related disciplines are independently distributed in different courses, and there is rarely a hands-on platform which includes signal processing, control theory, and artificial intelligence into Aerospace engineering. Facing this problem, this paper designs a multidisciplinary comprehensive experiment, aiming to provide a hand-on platform and flexible project-based program to students of aerospace engineering professions. First of all, in order to let the students understand actual aerospace problems, a multidisciplinary simulation platform containing UAVs and remote objects scenarios is constructed for them to explore in the experiments. Second, the content of the experiment is designed into three stages including data acquisition and processing, conceptual design and simulation, in-flight validation, during which the multidisciplinary engineering ability runs through the whole process of the activities. Finally, Project Oriented Design Based Learning is also introduced here to combine engineering design education with innovation and creativity. Through the project demonstration and presentation at the end of the experiment, the multidisciplinary engineering ability of each student can be effectively evaluated. The UVN comprehensive experiment enables students to work on real-world aerospace engineering problems through a hardware-software integration framework, which may greatly stimulate their curiosity and interest in autonomously learning. It also provides students unprecedented opportunities to immerse themselves in projects that cross disciplinary boundaries, improve their professional ability and enhance their exploration competence in aerospace areas.

This paper presents research that works in progress. Small private online courses (SPOCs) have received widespread attention for their adaptability to blended teaching in higher education. As an interactive tool, the SPOC discussion forum generates a large amount of data every day, including learning contents discussion, questions raising and feedback. In this paper, the computer structure course served as the research object, which is a SPOC for sophomores. Social network analysis (SNA) methods were utilized to explore the network extracted from the discussion forum. The results show that learners' three measures of centrality are significantly positively related to learning outcomes, and learners who play different roles in the discussion forum have a significant difference in their final grades. Our results can enable faculties to improve the curriculum and use online learning forums more effectively, such as increasing the number of teacher assistants (TAs) participating in the discussion forum, encouraging students to check the discussion forum regularly and post their learning feelings or questions.

Most educational institutions adopted the hybrid teaching mode through learning management systems. The logging data/clickstream could describe learners' online behavior. Many researchers have used them to predict students' performance, which has led to a diverse set of findings, but how to use insights from captured data to enhance learning engagement is an open question. Furthermore, identifying students at risk of failure is only the first step in truly addressing this issue. It is important to create actionable predictive model in the real-world contexts to design interventions. In this paper, we first extracted features from students' learning activities and study habits to predict students' performance in the Kung Fu style competency education. Then, we proposed a TrAdaBoost-based transfer learning model, which was pretrained using the data of the former course iteration and applied to the current course iteration. Our results showed that the generalization ability of the prediction model across the teaching iterations is high, and the model can achieve relatively high precision even when the new data are not sufficient to train a model alone. This work helped in timely intervention toward the at-risk students. In addition, two intervention experiments with split-test were conducted separately in Fall 2017 and Summer 2018. The statistical tests showed that both behavior-based reminding intervention and error-related recommending intervention that based on early prediction played a positive role in improving the blended learning engagement.