Department of Computer Science & Engineering

This report presents findings from a comprehensive survey of commercial fleet professionals involved in fleet safety (specifically managers, coaches and decision-makers) regarding the perceived effectiveness, practical implementation, and strategic value of different driver coaching approaches. Grounded in the broader objective of improving safety outcomes through evidence-based coaching strategies, the survey sought to explore both individual preferences and organizational practices.

The survey respondents consisted of two distinct groups: individuals who identify coaching as a key part of their primary job, and commercial dashcam decision-makers who do not directly engage in driver coaching. This composition allowed for insights from both those actively involved in coaching practices and those responsible for implementing or overseeing dashcam technology without hands-on coaching responsibilities.

This report critically evaluates coaching methodologies for enhancing fleet driver safety, engagement, and overall organizational performance. Drawing on empirical research across education, behavioral science, and fleet management, this analysis identifies key dimensions of effective commercial driver coaching, highlights significant limitations of current practices, and outlines strategic recommendations focused explicitly on differentiation and optimization. The findings emphasize the superior effectiveness of personalized, manager-led coaching methods, yet also show how manager-led coaching alone is challenging to implement at scale given the scarce availability of expert coaches. On the other hand, this report shows that in-cabin automated warning systems and self-coaching offer strengths in scalability and flexible learning. However, without integration with more personalized and interactive feedback, these approaches alone do not achieve the same long-term safety outcomes as when combined with human-led coaching elements. As such, this report recommends a mix of the three approaches for future optimization, especially focusing on (1) strategic integration of AI-driven platforms, and (2) complementing existing risk identification coaching programs with targeted positive reinforcement mechanisms and regular short-session, in-person coaching. Ultimately, the provided evidence-based recommendations offer a potential pathway for achieving measurable improvements in driver safety outcomes, operational efficiency, and competitive differentiation.

In this paper review, I will explain the fundamentals of a “laser soliton microcomb heterogeneously integrated on silicon”, presented by Xiang et. al of the Bowers Lab at UCSB [1]. I will try to relate concepts taught in class, which will be bolded, while presenting new material at a level that can be understood by someone who only has understanding of basic optics as well as the theory, components, and operation of a traditional Fabry-Perot cavity laser.

In this paper Xiang et. al present the design, the underlying photonics principles used, their fabrication process and considerations, and finally performance measurements and verification of the final device.

Background: Breastfeeding benefits both the mother and infant and is a topic of attention in public health. After childbirth, untreated medical conditions or lack of support lead many mothers to discontinue breastfeeding. For instance, nipple damage and mastitis affect 80% and 20% of US mothers, respectively. Lactation consultants (LCs) help mothers with breastfeeding, providing in-person, remote, and hybrid lactation support. LCs guide, encourage, and find ways for mothers to have a better experience breastfeeding. Current telehealth services help mothers seek LCs for breastfeeding support, where images help them identify and address many issues. Due to the disproportional ratio of LCs and mothers in need, these professionals are often overloaded and burned out.

Objective: This study aims to investigate the effectiveness of 5 distinct convolutional neural networks in detecting healthy lactating breasts and 6 breastfeeding-related issues by only using red, green, and blue images. Our goal was to assess the applicability of this algorithm as an auxiliary resource for LCs to identify painful breast conditions quickly, better manage their patients through triage, respond promptly to patient needs, and enhance the overall experience and care for breastfeeding mothers.

Methods: We evaluated the potential for 5 classification models to detect breastfeeding-related conditions using 1078 breast and nipple images gathered from web-based and physical educational resources. We used the convolutional neural networks Resnet50, Visual Geometry Group model with 16 layers (VGG16), InceptionV3, EfficientNetV2, and DenseNet169 to classify the images across 7 classes: healthy, abscess, mastitis, nipple blebs, dermatosis, engorgement, and nipple damage by improper feeding or misuse of breast pumps. We also evaluated the models' ability to distinguish between healthy and unhealthy images. We present an analysis of the classification challenges, identifying image traits that may confound the detection model.

Results: The best model achieves an average area under the receiver operating characteristic curve of 0.93 for all conditions after data augmentation for multiclass classification. For binary classification, we achieved, with the best model, an average area under the curve of 0.96 for all conditions after data augmentation. Several factors contributed to the misclassification of images, including similar visual features in the conditions that precede other conditions (such as the mastitis spectrum disorder), partially covered breasts or nipples, and images depicting multiple conditions in the same breast.

Conclusions: This vision-based automated detection technique offers an opportunity to enhance postpartum care for mothers and can potentially help alleviate the workload of LCs by expediting decision-making processes.

Prompts to large language models (LLMs) have evolved beyond simple user questions. For LLMs to solve complex problems, today's practices are to include domain-specific instructions, illustration of tool usages, and long context such as textbook chapters in prompts. As such, many parts of prompts are repetitive across requests, and their attention computation results can be reused. However, today's LLM serving systems treat every request in isolation, missing the opportunity of computation reuse.

This paper proposes Preble, the first distributed LLM serving platform that targets and optimizes for prompt sharing. We perform a study on five popular LLM workloads. Based on our study results, we designed a distributed scheduling system that co-optimizes computation reuse and load balancing. Our evaluation of Preble on two to 8 GPUs with real workloads and request arrival patterns on two open-source LLM models shows that Preble outperforms the state of the art avg latency by 1.5x to 14.5x and p99 by 2x to 10x.

Snapchat is one of the most popular social media apps in the world. It is no surprise, then, that many political ads are run on the service each year. Snap Inc.'s political ads library is part of an effort by the company to increase transparency in their advertising practices. The data analyzed in this project spans 2019-2020, and consists of information on every political ad that was run on the service in that timeframe, including who the ad buyer was, how much the ad cost, what areas it targeted, etc. Geographic and monetary distribution of ads is analyzed and possible explanations given for anomalies. Missingness of the data was evaluated and Vermont was identified as an area with unusual spending. With α = 0.05 the null hypothesis was rejected (p = 0.02); the distribution of ad dollars to Vermont is not wholly random.

The Snapchat ads dataset contains political ad data for ads on Snapchat, oneof the largest social media networks in the world. A key feature of the datasetis how much money an organization spends on a particular ad, found in the`Spend` column. It is reasonable to assume that this amount varies based oncertain factors, but can we use those factors to figure out how much is spenton an ad? We can explore this by predicting ad spending through machinelearning. After feature analysis and engineering, we arrive at a linearregression model with $R^2$ = .85 and perform a fairness evaluation of thealgorithm.

Digital currencies have flourished in recent years, buoyed by the tremendous success of Bitcoin. These blockchain-based currencies, called altcoins, have attracted enthusiasts who enter the market by mining or buying them. To mine or to buy, however, can be a difficult decision; each altcoin is different from another, and the market tends to be volatile. In this work, we analyze the profitability of mining and speculation for 36 altcoins using real-world blockchain and trade data. Using opportunity cost as a metric, we estimate the mining cost for a coin with respect to a more popular coin. For every dollar invested in mining or buying a coin, we also estimate the revenue under various conditions, such as time of market entry and hold positions. While some coins offer the potential for spectacular returns, many follow a simple bubble-and-crash scenario, which highlights the extreme risks---and potential gains---in altcoin markets.

Pre-2018 CSE ID: CS2017-1019

Emerging fast, persistent memories will enable systems that combine conventional DRAM with large amounts of non-volatile main memory (NVMM) and provide huge increases in storage performance. Fully realizing this potential requires fundamental changes in how system software manages, protects, and provides access to data that resides in NVMM. We address these needs by describing a NVMM-optimized file system called NOVA that is both fast and resilient in the face of corruption due to media errors and software bugs. We identify and propose solutions for the unique challenges in hardening an NVMM file system, adapt state-of-the-art reliability techniques to an NVMM file system, and quantify the performance and storage overheads of these techniques. We find that NOVA's reliability features increase file system size system size by 14.9% and reduce application-level performance by between 2% and 38%.

Pre-2018 CSE ID: CS2017-1018

In this paper we introduce Echidna, a hybrid schematic/ text-based language for describing PCB circuit schematics. Echidna allows designers to use high-level programming con- structs to describe schematics, supports modular, reusable design components with well-defined interfaces, and provides for complex parameterization of those modules. Echidna deeply integrates a high-level programming language into a schematic-based design flow. The designer can describe schematics in code, as a schematic, or as a seamless combination of the two. We demonstrate its usefulness with several case studies.

Pre-2018 CSE ID: CS2016-1017