Search for: All records

Objectives The traditional, instructor-led form of live coding has been extensively studied, with findings showing that this form of live coding imparts similar learning to static-code examples. However, a concern with Traditional Live Coding is that it can turn into a passive learning activity for students as they simply observe the instructor program. Therefore, this study compares Active Live Coding—a form of live coding that leverages in-class coding activities and peer discussion—to Traditional Live Coding on three outcomes: 1) students’ adherence to effective programming processes, 2) students’ performance on exams and in-lecture questions, and 3) students’ lecture experience. Participants Roughly 530 students were enrolled in an advanced, CS1 course taught in Java at a large, public university in North America. The students were primarily first- and second-year undergraduate students with some prior programming experience. The student population was spread across two lecture sections—348 students in the Active Live Coding (ALC) lecture and 185 students in the Traditional Live Coding (TLC) lecture. Study Methods We used a mixed-methods approach to answer our‘ research questions. To compare students’ programming processes, we applied process-oriented metrics related to incremental development and error frequencies. To measure students’ learning outcomes, we compared students’ performance on major course components and used pre- and post-lecture questionnaires to compare students’ learning gain during lectures. Finally, to understand students’ lecture experience, we used a classroom observation protocol to measure and compare students’ behavioral engagement during the two lectures. We also inductively coded open-ended survey questions to understand students’ perceptions of live coding. Findings We did not find a statistically significant effect of ALC on students’ programming processes or learning outcomes. It seems that both ALC and TLC impart similar programming processes and result in similar student learning. However, our findings related to students’ lecture experience shows a persistent engagement effect of ALC, where students’ behavioral engagement peaks andremains elevatedafter the in-class coding activity and peer discussion. Finally, we discuss the unique affordances and drawbacks of the lecture technique as well as students’ perceptions of ALC. Conclusions Despite being motivated by well-established learning theories, Active Live Coding did not result in improved student learning or programming processes. This study is preceded by several prior works that showed that Traditional Live Coding imparts similar student learning and programming skills as static-code examples. Though potential reasons for the lack of observed learning benefits are discussed in this work, multiple future analyses to further investigate Active Live Coding may help the community understand the impacts (or lack thereof) of the instructional technique. 

Supervised learning of skeleton sequence encoders for action recognition has received significant attention in recent times. However, learning such encoders without labels continues to be a challenging problem. While prior works have shown promising results by applying contrastive learning to pose sequences, the quality of the learned representations is often observed to be closely tied to data augmentations that are used to craft the positives. However, augmenting pose sequences is a difficult task as the geometric constraints among the skeleton joints need to be enforced to make the augmentations realistic for that action. In this work, we propose a new contrastive learning approach to train models for skeleton-based action recognition without labels. Our key contribution is a simple module, HaLP – to Hallucinate Latent Positives for contrastive learning. Specifically, HaLP explores the latent space of poses in suitable directions to generate new positives. To this end, we present a novel optimization formulation to solve for the synthetic positives with an explicit control on their hardness. We propose approximations to the objective, making them solvable in closed form with minimal overhead. We show via experiments that using these generated positives within a standard contrastive learning framework leads to consistent improvements across benchmarks such as NTU-60, NTU- 120, and PKU-II on tasks like linear evaluation, transfer learning, and kNN evaluation. Our code can be found at https://github.com/anshulbshah/HaLP. 

Incremental development is the process of writing a small snippet of code and testing it before moving on. For students in introductory programming courses, the value of incremental development is especially higher as they may suffer from more syntax errors, lack the proficiency to address complicated bugs, and may be more prone to frustration when struggling to correct code. However, to evaluate the effectiveness of interventions that aim to teach programming processes such as incremental development, we need to develop measures to assess such processes. In this paper, we present a way to measure incremental development. By qualitatively analyzing 15 student coding interviews, we identified common behaviors in the programming process that relate to incremental development. We then leveraged a dataset of over 1000 development sessions -- about 52,000 code snapshots at compilation time -- to automatically detect the common behaviors identified in our qualitative analysis. Finally, we crafted a formal metric, called the ``Measure of Incremental Development’' (MID), to quantify how effectively a student used incremental development during a programming session. The MID detects common non-incremental development patterns such as excessive debugging after large additions of code to automatically assess a sequence of snapshots. The MID aligns with human evaluations of incrementality with over 80% accuracy. Our metric enables new research directions and interventions focused on improving students' development practices. 

Standard contrastive learning approaches usually require a large number of negatives for effective unsupervised learning and often exhibit slow convergence. We suspect this behavior is due to the suboptimal selection of negatives used for offering contrast to the positives. We counter this difficulty by taking inspiration from support vector machines (SVMs) to present max-margin contrastive learning (MMCL). Our approach selects negatives as the sparse support vectors obtained via a quadratic optimization problem, and contrastiveness is enforced by maximizing the decision margin. As SVM optimization can be computationally demanding, especially in an end-to-end setting, we present simplifications that alleviate the computational burden. We validate our approach on standard vision benchmark datasets, demonstrating better performance in unsupervised representation learning over state-of-the-art, while having better empirical convergence properties.