Traffic Engineering (TE) in large-scale networks like cloud Wide Area Networks (WANs) and Low Earth Orbit (LEO) satellite constellations is a critical challenge. Although learning-based approaches have been proposed to address the scalability of traditional TE algorithms, their practical application is often hindered by a lack of generalization, high training overhead, and a failure to respect link capacities. This paper proposes TELGEN, a novel TE algorithm that learns to solve TE problems efficiently in large-scale network scenarios, while achieving superior generalizability across diverse network conditions. TELGEN is based on the novel idea of transforming the problem of “predicting the optimal TE solution” into “predicting the optimal TE algorithm”, which enables TELGEN to learn and efficiently approximate the end-to-end solving process of classical optimal TE algorithms. The learned algorithm is agnostic to the exact underlying network topology or traffic patterns, and is able to very efficiently solve TE problems given arbitrary inputs and generalize well to unseen topologies and demands. We train and evaluate TELGEN with random and real-world topologies, with networks of up to 5000 nodes and 3.6×106links in testing. TELGEN shows less than 3% optimality gap while ensuring feasibility in all testing scenarios, even when the test network has 2-20× more nodes than the largest training network. It also saves up to 84% TE solving time than traditional interior-point method, and reduces up to 79.6% training time per epoch than the state-of-the-art learning-based algorithm.

A technical interview is the culmination of the recruiting process for hiring software engineers in the tech industry. Many well-known companies, including Amazon, Meta (formerly Facebook), Alphabet (Google), and Microsoft, use it to filter candidates. However, the drawbacks of technical interviews are well-documented, including their lack of real-world relevance, bias towards newer developers, demanding time commitment, and potential to induce unnecessary anxiety and frustration. De-spite these criticisms, there is no clear indication that the industry will alter the format of technical interviews in the near future. To assist student developers in preparing for these challenges, we conducted a quantitative analysis using over 300,000 user profiles from LeetCode, arguably the most popular online platform for preparing software development candidates for interviews. Our analysis aims to provide developers with insights into the effort required to prepare for technical interviews, especially in terms of solving programming questions, to secure a position at a renowned company.

This is a research-to-practice full paper. Several research studies indicate that women who have opted into a computing career path must regularly contend with negative stereotypes about their technical abilities. These stereotypes are often cited as contributing factors to the underrepresentation of women in computing. To counter these stereotypes and enhance female participation in computer science, numerous interventions have been designed. However, most existing research tends to rely on anecdotal evidence and questionnaires to study these stereotypes. In contrast, our study collected data from over 900 students over a span of eight years and adopted a comprehensive quantitative approach to examine these stereotypes about female students. We utilized pre-class GitHub contribution metrics to evaluate students' programming experience and an array of in-class grading items to measure students' performance. Additionally, we mined the project repositories' git logs to gain insights into students' contributions to team projects. Our investigation began by probing whether there was a notable difference in the technical backgrounds or preparedness between female and male students. The results indicated that males tended to be better prepared. Next, we explored potential disparities in class performance between the two genders. Our findings revealed that males and females each excelled in different areas. We were also interested in discerning if female and male students contributed equally to team projects; our analysis affirmed that the contributions were comparable between the two groups. If allowed to choose their teammates, we examined whether they showed a preference for single-gender teams or mixed-gender teams. Our conclusions indicated no marked preference. This paper aims to augment the body of research on computing education by assisting educators in gaining a better understanding of female students in the class. Moreover, it tests the stereotypes by comparing them with empirical results.

In modern software engineering education, team formation is crucial for mimicking real-world collaborative scenarios and boosting project-based learning outcomes. This paper introduces a simple, innovative, and universally adaptable method for forming student teams within a software engineering class. We utilize publicly available pre-class GitHub metrics as our input variables (e.g., number of commits, pull requests, code size, etc.). For team formation, the constrained k-means algorithm is employed. This algorithm embraces domain-specific constraints, ensuring the resulting teams not only resonate with the inherent data clusters but also meet educational requirements. Preliminary results suggest that our methodology yields teams with a harmonious blend of skills, experiences, and collaborative potentials, thereby setting the stage for enhanced project success and enriched learning experiences. Quantitative analyses show that teams formed via our approach outperform both randomly assembled teams and student self-selected teams concerning project grades. Moreover, teams created using our method also display a reduced standard deviation in grades, suggesting a more consistent performance across the board.

This study presents a comparative analysis of contributions to GitHub by students before and after participating in a Software Engineering class based on Open Source Software (OSS). The primary objective is to understand the influence of formal software engineering education on students' engagement in OSS projects, as reflected in their GitHub activities. The research addresses two key questions. Firstly, it examines how GitHub contributions change before and after the class. The corresponding hypothesis posits that students' average GitHub contributions will exhibit a distinct pattern post-class compared to pre-class. Additionally, the study explores the potential association between students' academic performance in the class and their level of GitHub contributions after the class. The strength and direction of the potential association are quantified using the Spearman correlation coefficient, considering the potential non-linear nature of the data. This analysis uses data from over 1000 students across more than 10 years, encompassing their GitHub contribution data over multiple timeframes and their grades in the class. The study employs a combination of statistical methods, including paired tests and correlation analysis, to explore these dynamics. While causality cannot be established due to the absence of a control group, the findings offer valuable insights into the correlation between academic engagement and practical contributions in the realm of OSS development. This research contributes to the understanding of how theoretical software engineering education might relate to practical application and engagement in real-world projects.

Group or team projects are an essential component of the software engineering curriculum. Earlier studies have explored how prior programming experience influences students' team project performance and overall class performance in software engineering. However, few studies address the impact of prior programming experience on students' contributions to team projects. Previous work has varied in its definitions of prior programming experience or skill, leading to inconsistent findings. In this study, we collected pre-class GitHub contribution metrics from 237 students (forming 79 teams of three) across two academic years to measure their prior programming experience and skills. We also mined students' project repositories' git logs to collect individual student contributions. A central question revolved around whether students with more substantial prior programming experience were indeed more active contributors to their project teams. Interestingly, our data indicated a positive correlation between prior programming experience and contributions to team projects. We further delved into team dynamics. Specifically, we questioned if teams made up of members with comparable skill levels exhibited a more even distribution of contributions. Contrary to expectations, our findings revealed no association between these two variables. Moreover, we investigated the team configurations that might encourage the rise of "free riders"-students who contributed only minimally. This paper seeks to augment the body of research on computing education and assist educators in understanding how prior programming experience impacts students' contributions in team projects.

To enable the inference of high-precision deep neural networks (DNNs) on resource-constrained devices, DNN offloading has been widely explored in recent years. Some works have also integrated the chain-topology DNN (CDNN) offloading with pipeline processing to further reduce inference delay when processing streaming tasks. To improve the accuracy of the inference results, the topology of DNN tends to evolve from chain topology to directed acyclic graph (DAG) topology. However, most of the existing works do not study partitioning and offloading DAG-topology DNNs (DDNNs) for streaming tasks. Moreover, when partitioning computationally expensive DNN models, multipartitioning probably outperforms the bi-partitioning method, and most of the works do not study multipartitioning DAG-topology DNNs. In this article, we propose a more general multipartitioning and offloading method for large-scale DDNNs to process streaming tasks, which can adaptively partition DDNNs into multiple parts considering the computing power and bandwidth of all available computing units. Specifically, we first present a transforming method based on topological sorting that can losslessly transform DAG-topology DNNs into CDNNs. Then, based on greedy and dichotomy ideas, a multipartitioning algorithm is designed to partition and offload CDNNs. In this way, we can solve DDNNs’ multipartitioning problem based on the proposed transforming and partitioning algorithms. Experimentshttps://github.com/sreasearcher/PDD-Codeshow that the method proposed in this article significantly outperforms bi-partitioning and nonpartitioning methods when offloading computationally expensive DNN models.

A quantum network distributes quantum entanglements between remote nodes, and is key to many applications in secure communication, quantum sensing and distributed quantum computing. This paper explores the fundamental trade-off between the throughput and the quality of entanglement distribution in a multi-hop quantum repeater network. Compared to existing work which aims to heuristically maximize the entanglement distribution rate (EDR) and/or entanglement fidelity, our goal is to characterize the maximum achievable worst-case fidelity, while satisfying a bound on the maximum achievable expected EDR between an arbitrary pair of quantum nodes. This characterization will provide fundamental bounds on the achievable performance region of a quantum network, which can assist with the design of quantum network topology, protocols and applications. However, the task is highly non-trivial and is NP-hard as we shall prove. Our main contribution is a fully polynomial-time approximation scheme to approximate the achievable worst-case fidelity subject to a strict expected EDR bound, combining an optimal fidelity-agnostic EDR-maximizing formulation and a worst-case isotropic noise model. The EDR and fidelity guarantees can be implemented by a post-selection-and-storage protocol with quantum memories. By developing a discrete-time quantum network simulator, we conduct simulations to show the characterized performance region (the approximate Pareto frontier) of a network, and demonstrate that the designed protocol can achieve the performance region while existing protocols exhibit a substantial gap.

Scalability is a critical challenge for blockchain-based cryptocurrencies. Payment channel networks (PCNs) have emerged as a promising solution for this challenge. However, channel balance depletion can significantly limit the capacity and usability of a PCN. Specifically, frequent transactions that result in unbalanced payment flows from two ends of a channel can quickly deplete the balance on one end, thus blocking future payments from that direction. In this paper, we propose Fence, an online balance-aware fee setting algorithm to prevent channel depletion and improve PCN sustainability and long-term throughput. In our algorithm, PCN routers set transaction fees based on the current balance and level of congestion on each channel, in order to incentivize payment senders to utilize paths with more balance and less congestion. Our algorithm is guided by online competitive algorithm design, and achieves an asymptotically tight competitive ratio with constant violation in a unidirectional PCN. We further prove that no online algorithm can achieve a finite competitive ratio in a general PCN. Extensive simulations under a real-world PCN topology show that Fence achieves high throughput and keeps network channels balanced, compared to state-of-the-art PCN routing algorithms.

Research into predicting students' performance in computer science classes has been conducted globally for over five decades. Numerous metrics, including performance in prior courses, demographic information, and programming experience, have been used to predict success in computer science. Various analytical methods, such as linear regression, decision trees, ensemble methods, and even neural networks, have also been explored. In this study, we investigate whether pre-class GitHub contribution metrics, combined with machine learning techniques, can forecast student performance in a software engineering class. We address two research questions in this paper. Firstly, can pre-class GitHub contribution metrics predict students' performance? Secondly, which machine learning technique is most effective in predicting student performance? We collected data from 802 students over five years and 11 semesters, including pre-class GitHub contribution stats, students' exam grades, project grades, documentation grades, and review writing grades. Eight different machine learning methods were then tested to predict in-class performance using pre-class GitHub contributions. Our results indicate that exam performance can be relatively accurately predicted by machine learning methods. Ensemble methods such as Random Forest, AdaBoost, and XGBoost performed better than other methods. This suggests that pre-class GitHub contribution metrics can be a useful tool for predicting students' performance in software engineering classes, carrying significant implications for educators. This approach can help educators identify at-risk students at the earliest point in the class, enabling early intervention strategies to prevent failure. Our study uniquely utilizes pre-class GitHub contributions, providing a preliminary indication of a student's familiarity with the course material. While prior research has focused on using in-class data to predict student performance, our approach identifies struggling students from the very beginning. We believe this can provide the most beneficial support for students.

Communication bottleneck and statistical heterogeneity are two critical challenges of federated learning (FL) over wireless networks. To tackle both challenges, in this paper we propose an over-the-air computation (AirComp) assisted hierarchical personalized FL (HPFL) framework, where a device-edge-cloud based three-tier network architecture is adopted to simultaneously learn a global model and multiple personalized local models. We analyze the convergence of the AirComp-assisted HPFL framework and formulate an optimization problem to minimize the transmission distortion, which is an essential component of the convergence upper bound. An efficient algorithm is subsequently developed to optimize the transceiver design by leveraging successive convex approximation and Lagrangian duality. We conduct extensive simulations to demonstrate that our developed algorithm achieves a near-optimal performance and a much greater test accuracy than the baseline algorithms.

Quantum entanglement distribution between remote nodes is key to many promising quantum applications. Existing mechanisms have mainly focused on improving throughput and fidelity via entanglement routing or single-node scheduling. This paper considers entanglement scheduling and distribution among many source-destination pairs with different requests over an entire quantum network topology. Two practical scenarios are considered. When requests do not have deadlines, we seek to minimize the average completion time of the communication requests. If deadlines are specified, we seek to maximize the number of requests whose deadlines are met. Inspired by optimal scheduling disciplines in conventional single-queue scenarios, we design a general optimization framework for entanglement scheduling and distribution called ESDI, and develop a probabilistic protocol to implement the optimized solutions in a general buffered quantum network. We develop a discrete-time quantum network simulator for evaluation. Results show the superior performance of ESDI compared to existing solutions.

The wide spread of vehicular cameras has raised broad privacy concerns. Ubiquitous vehicular cameras capture bystanders like people or cars nearby without their awareness. To address privacy concerns, most existing works either blur out direct identifiers such as vehicle license plates and human faces, or obfuscate whole video frames. However, the former solution is vulnerable to re-identification attacks based on general features, and the latter severely impacts utility of the transformed videos. In this paper, we propose an INStance-level PrIvacy-pREserving (INSPIRE) video transformation framework for vehicular camera videos. INSPIRE leverages deep neural network models to detect and replace sensitive object instances in vehicular videos with their non-existent counterparts. We design INSPIRE as a modular framework to enable flexible customization of protected instance categories and their protection modules. An implementation of INSPIRE focused on protecting people and cars is described, which we tested on six re-identification datasets and three real-world vehicular video datasets to evaluate its privacy protection and utility preservation capability. Results show that INSPIRE can thwart 97% of re-identification attacks for people and cars while maintaining a 0.75 object detection mean average precision on transformed instances. We also demonstrate experimentally that INSPIRE is robust against model inversion attacks. Compared to solutions that provide comparable privacy protection, INSPIRE achieves relatively 1.76 times higher counting accuracy and 31.61% higher object detection mean average precision.

Edge computing promises to bring low-latency and high-throughput computing, but the limited edge resources may cause frequent congestion and lead to unstable and unpredictable performance. To ensure performance guarantee, application owners can establish Service-Level Agreements (SLAs) with the edge provider for resource reservation or priority usage. But it is cost-inefficient for application owners to lease long-term SLAs based on peak demands, as demands can fluctuate, and the leased resources may be idle or underutilized at most times. This paper studies market mechanism design for short-term edge SLA leases, focusing on real-time big data applications with throughput and latency goals. Applications submit short-term SLA requests to serve users with guaranteed performance during peak hours. As SLA requests arrive over time, the edge provider dynamically provisions edge resources to fulfill the requests, while charging application owners based on the current demands. We design EA-Market, an online combinatorial auction mechanism that achieves a competitive social welfare, while guaranteeing truthfulness, budget balance, individual rationality, and computational efficiency. Notably, our mechanism enables each application owner to bid without knowledge of the edge infrastructure, and gives edge provider full control over resource provisioning to fulfill the requests. We perform theoretical analysis and simulations to evaluate the efficacy of our mechanism.

What skills does a student need to succeed in a programming class? Ostensibly, previous programming experience may affect a student's performance. Most past studies on this topic use self-reporting questionnaires to query students about their programming experience. This paper presents a novel, unified, and replicable way to measure previous programming experience using students' pre-class GitHub contributions. To our knowledge, we are the first to use GitHub contributions in this way. We conducted a comprehensive statistical study of students in an object-oriented design and development class from 2017 to 2022 (n = 751) to explore the relationships between GitHub contributions (commits, comments, pull requests, etc.) and students' performance on exams, projects, designs, etc. in the class. Several kinds of contributions were shown to have statistically significant correlations with performance in the class. A set of two-samplet -tests demonstrate statistical significance of the difference between the means of some contributions from the high-performing and low-performing groups.

Deploying federated learning (FL) in wireless networks faces the critical challenge of communication bottlenecks. To address this issue, in this paper, we consider an over-the-air computation (AirComp) assisted hierarchical FL (HFL) framework, where a cloud-edge-device-based three-tier network architecture is constructed to train a global model. We first theoretically characterize the convergence of the AirComp-assisted HFL framework and formulate a combinatorial optimization problem that jointly optimizes the edge interval control and local device transceiver design to minimize the convergence upper bound to boost the overall learning performance and reduce communication cost. We show that the formulated optimization problem can be decoupled into an edge interval control problem and a transceiver design problem, which can be tackled by developing a relaxation and rounding algorithm and an alternating Lyapunov drift-based algorithm, respectively. Extensive simulations demonstrate that our proposed algorithm significantly outperforms the baseline schemes.

This paper studies how an edge-based federated learning algorithm called FedAegis can be designed to be ro-bust under both heterogeneous data distributions and Byzantine adversaries. The divergence of local data distributions leads to suboptimal results for the training process of federated learning, and the Byzantine adversaries aim to prevent the training process from converging in a distributed learning system. In this paper, we show that an edge-based hierarchical federated learning architecture can help tackle this dilemma by utilizing edge nodes geographically close to clusters of local devices. By combining a distributionally robust global loss function with a local Byzantine-robust aggregation rule, FedAegis can defend against remote Byzantine adversaries who cannot manipulate local devices' connections to edge nodes, meanwhile accounting for global data heterogeneity across benign local devices. Experiments with the MNIST, FMNIST and CIFAR-IO datasets show that our proposed algorithm can achieve convergence and high accuracy under heterogeneous data and various attack scenarios, while state-of-the-art defenses and robustness mechanisms are non-converging or have reduced average and/or worst-case accuracy.

Payment Channel Network (PCN) is an auspicious solution to the scalability issue of the blockchain, improving transaction throughput without relying on on-chain transactions. In a PCN, nodes can set prices for forwarding payments on behalf of other nodes, which motivates participation and improves network stability. Analyzing the price setting behaviors of PCN nodes plays a key role in understanding the economic properties of PCNs, but has been under-studied in the literature. In this paper, we apply equilibrium analysis to the price-setting game between two payment hubs in the PCN with limited channel capacities and partial overlap demand. We analyze existence of pure Nash Equilibriums (NEs) and bounds on the equilibrium revenue under various cases, and propose an algorithm to find all pure NEs. Using real data, we show bounds on the price of anarchy/stability and average transaction fee under realistic network conditions, and draw conclusions on the economic advantage of the PCN for making payment transfers by cryptocurrency users.

This paper studies how to provision edge computing and network resources for complex microservice-based applications (MSAs) in face of uncertain and dynamic geo-distributed demands. The complex inter-dependencies between distributed microservice components make load balancing for MSAs extremely challenging, and the dynamic geo-distributed demands exacerbate load imbalance and consequently congestion and performance loss. In this paper, we develop an edge resource provisioning model that accurately captures the inter-dependencies between microservices and their impact on load balancing across both computation and communication resources. We also propose a robust formulation that employs explicit risk estimation and optimization to hedge against potential worst-case load fluctuations, with controlled robustness-resource trade-off. Utilizing a data-driven approach, we provide a solution that provides risk estimation with measurement data of past load geo-distributions. Simulations with real-world datasets have validated that our solution provides the important robustness crucially needed in MSAs, and performs superiorly compared to baselines that neglect either network or inter-dependency constraints.