As machine learning (ML) becomes integral in more disciplines, introductory courses in the field are attracting increasingly diverse audiences. Design of these introductory ML courses needs to be theoretically sound, but also intuitive, engaging, and accessible to a range of students. Effective teaching of ML must go beyond teaching the theoretical or practical mechanics of algorithms. In this paper, we synthesize effective teaching strategies from 6 experienced ML instructors across 5 institutions to help students define appropriate ML problems, build intuition, develop reasoning skills, and apply models responsibly. We organize these strategies into eight thematic areas: preparing students for success, motivating learners through real-world relevance, integrating ethics and societal impact, avoiding common methodological pitfalls in model evaluation, guiding students on design decisions, adapting effective classroom practices, assessing student learning, and preparing for the future. Each section offers practical examples of classroom-tested activities (or references to existing resources), and in many cases, reflections on our experiences with the strategies. Our aim is for this paper to be a starting point for instructors aiming to improve learning in introductory ML courses. We hope this is a resource-rich guide for teaching ML to diverse learners, grounded in both pedagogy and practice.

Machine learning is increasingly used across wireless systems, with research papers reporting highly promising results, yet most proposed solutions are never deployed in practice. One major reason is the generalization gap between research evaluation and real world wireless settings. In this paper, we reproduce studies that apply machine learning to wireless systems across three application areas: throughput prediction, channel estimation, and sensing from wireless signals. We examine how data collection, processing, and splitting decisions affect reported performance. Across these case studies, we find that common evaluation practices, such as random assignment of samples to training and test sets, can introduce relationships between training and test sets that would not exist in the proposed deployment setting. These arise from temporal, environmental, or subject correlations in wireless measurements and can lead to overly optimistic estimates of model performance. When we redesign the evaluation to better reflect intended deployment settings, the estimated model performance is much worse, revealing substantial generalization gaps. We conclude with practical recommendations for designing and reporting machine learning evaluations for wireless systems, emphasizing data partitioning strategies that reflect the intended deployment setting.

Internet switches become harder and costlier to build for higher line rates and switch capacities. In-sequence delivery of packets has traditionally been a constraint on switch designs because TCP loss detection was considered vulnerable to out-of-sequence arrivals. For this reason, extremely efficient and simple designs, such as the Load Balanced Birkhoff-von Neumann Switch, were considered impractical. However, we reevaluate this constraint considering modern TCP implementations with loss detection algorithms like Recent Acknowledgment (RACK) that are more resilient to out-of-order arrivals. In a set of testbed experiments representative of wide area core networks, we evaluated the performance of TCP flows traversing a load balanced switch that reorders some packets within a flow. We show that widely deployed and standard TCP implementations of the last decade achieve similar performance when traversing a load balanced switch as they do when there is no reordering. Furthermore, we also verified that an increase in the line rate leads to favorable conditions for time based loss detection methods, such as the one used in RACK. Our results, if further validated, suggest that switch designs that were previously thought to be unsuitable can potentially be utilized, thanks to the relaxation of the in-sequence delivery constraint.

We replicate the paper, "When to Use and When Not to Use BBR: An Empirical Analysis and Evaluation Study" by Cao et al, published in IMC 2019 [2], with a focus on the relative goodput of TCP BBR and TCP CUBIC for a range of bottleneck buffer sizes, bandwidths, and delays. We replicate the experiments performed by the original authors on two large-scale open-access testbeds, to validate the conclusions of the paper. We further extend the experiments to BBRv2. We package the experiment artifacts and make them publicly available so that others can repeat and build on this work.

PrairieLearn is an open source, extensible online assessment platform built on modern web technologies. In this workshop, we will focus on how PrairieLearn can be used to improve student learning in undergraduate computer science classes. However, the platform is also more than suitable for use as an assessment engine in a variety of courses including the humanities, social, physical, and life sciences. In the first part of the workshop, we will showcase multiple question styles that highlight PrairieLearn's abilities as an online platform, including deploying automatically and manually graded questions at scale in large classes. In the second part of the workshop, we will discuss the anatomy of a PrairieLearn question, create several custom questions, and design assessments in PrairieLearn. In the third part, we will share strategies on adopting PrairieLearn at your institution. In particular, how algorithmically generated questions can be used in support of alternative grading schemes such as Mastery- or Specifications-Grading. Finally, we will share how PrairieLearn can be extended to support other coding languages and paradigms with custom and external autograders. There will be plenty of opportunities for questions throughout the workshop, and we intend to leave plenty of time for additional 1:1 support and training. Attendees will be able to attend the session virtually and are recommended to bring a web-connected computing device. By the end of the session, attendees will know enough to run a whole class on PrairieLearn including designing questions appropriate for homework, labs, and tests.

Next generation cellular networks will rely heavily on mmWave and THz spectrum for the abundantly available bandwidth. However, these frequencies suffer from high path and penetration losses. One way to improve the performance is network densification which comes with higher operational cost. To reduce the operational cost of Base Station (BS) deployment, the 3GPP has proposed the Integrated Access and Backhaul (IAB) architecture. Nonetheless, when a handover does occur in IAB networks, even with minimal handover time, the traffic that is already en route to the UE is delayed, as the new serving BS must retrieve the packets from either the previous serving BS or the core. To address these challenges, we propose Fast Wireless Backhaul (FWB), a new wireless backhaul solution. FWB takes advantage of the multi-connectivity of UEs and the high-capacity low-cost wireless backhaul promised by IAB to reduce latency and increase reliability in case of unexpected blockages on the wireless signal path. In FWB, the BSs serving the UE participate in a multicast tree, and receive all packets designated for the UE, but only one BS transmits packets to the UE. In the event of a link failure between the UE and the serving BS, another BS takes over to maintain the data plane connection, without first having to retrieve undelivered downlink packets. We believe our architecture can enable mission critical applications with stringent latency and reliability requirements.

With the new emerging throughput-intensive ultralow latency applications, there is a need for a transport layer protocol that can achieve high throughput with low latency. One promising candidate is TCP BBR, a protocol developed by Google, with the aim of achieving high throughput and low latency by operating around the Bandwidth Delay Product (BDP) of the bottleneck link. Google reported significant throughput gains and much lower latency relative to TCP Cubic following the deployment of BBR in their high-speed wide area wired network. As most of these emerging applications will be supported by Millimeter Wave (mmWave) wireless networks, BBR should achieve both high throughput and ultra-low latency in these settings. However, in our preliminary experiments with BBR over a mmWave wireless link operating at 60 GHz, we observed a severe degradation in throughput that we were able to attribute to high delay variation on the link. In this paper, we show that “throughput collapse” occurs when BBR's estimate of minimum RTT is less than half of the average RTT of the uncongested link (as when delay jitter is large). We demonstrate this phenomenon and explain the underlying reasons for it using a series of controlled experiments on the CloudLab testbed. We also present a mathematical analysis of BBR, which matches our experimental results closely. Based on our analysis, we propose and experimentally evaluate potential solutions that can overcome the throughput collapse without addina sianificant latency.

With the increasing interest in the use of millimeter wave bands for 5G cellular systems comes renewed interest in resource sharing. Properties of millimeter wave bands such as massive bandwidth, highly directional antennas, high penetration loss, and susceptibility to shadowing, suggest technical advantages to spectrum and infrastructure sharing in millimeter wave cellular networks. However, technical advantages do not necessarily translate to increased profit for service providers, or increased consumer surplus. In this paper, detailed network simulations are used to better understand the economic implications of resource sharing in a vertically differentiated duopoly market for cellular service. The results suggest that resource sharing is less often profitable for millimeter wave service providers compared to microwave cellular service providers, and does not necessarily increase consumer surplus.

The resource usage model for millimeter wave bands has been the subject of considerable debate. The massive bandwidth, highly directional antennas, high penetration loss and susceptibility to shadowing in these bands suggest certain advantages to spectrum and infrastructure sharing. In particular, resources that are "open", such as unlicensed spectrum or a deployment of base stations open to all service providers, may offer greater gains in mmWave bands than at conventional cellular frequencies. However, even when sharing is technically beneficial (as recent research in this area suggests that it is), it may not be profitable. In this paper, both the technical and economic implications of resource sharing in millimeter wave networks are studied. Millimeter wave service is considered in the economic framework of a network good, and detailed network simulations are used to understand data rates, profit, and demand for millimeter wave service, with and without open resources. The results suggest that "open" deployments of neutral small cells that serve subscribers of any service provider encourage market entry by making it easier for networks to reach critical mass, more than "open" (unlicensed) spectrum would.

A common simplifying assumption made in wireless simulation and modeling is that the world is flat, i.e. to ignore the effect of the terrain in which the wireless signal propagates. In this paper, we show with empirical measurements from an urban wireless network testbed how the terrain affects the spatial and temporal correlation of the wireless signal, and in turn, the distance or duration over which the wireless signal remains consistent. Furthermore, we suggest that this effect has practical implications for systems that make assumptions about the duration over which wireless signal quality stays roughly the same, such as adaptive transmission schemes or applications that buffer data to smooth over variations in signal quality.