2021-2022 Projects

Listed alphabetically by project sponsor.

Allthenticate

Quantifying Security - Visualizing Usability of Allthenticate

Student team: Akshat Ataliwala, Calvin Jenkins, Bosco De Enrique Romeu, and Yuki Yamazaki

Project advisors: Chad Spensky and Leron Reznikov

Allthenticate provides a proximity-based app-enabled locking/unlocking system in place of traditional badge or lock-and-key security systems. The primary objectives of this project are to quantify the benefits that our Allthenticate’s security product provides and identifying opportunities to improve the Allthenticate user experience.

First, we want to quantify how much of an improvement Allthenticate’s security solution provides over existing solutions. This includes tasks such as measuring time saved by using Allthenticate, tracking changes in Bluetooth connection, and monitoring app status. We will analyze the user data provided by the logs from readers, as well as the phone and computer apps. Second, we intend to increase how well the product is working. Allthenticate is constantly improving, and with user-happiness being at the core of the company’s values, we want to cater the best experience for the user. An example of this could be predicting a user’s app behaviors to personalize their UI to give them the most seamless experience for their particular needs.

Using preliminary product data, we developed a visualization dashboard for real-time monitoring of a client’s system activity.

AppFolio

Deep Learning Recommendation Systems for Property Management Software

Student team: Alyssa Keehan, Jordan Tran, Joshua Harasaki, Philip Carey, and Romina Fareghbal

Project advisors: Shyr-Shea Chang, Soeren Thust, and Erika McPhillips

For this project, we created a recommender system that predicts users interests and recommends properties to users based on previous user interactions and property features. Recommender systems are among the most powerful machine learning systems that are used to help users identify items they may be interested in and also helps companies increase interaction and sales. Our sponsor is AppFolio, a tech company based in Santa Barbara that provides innovative software, services, and data analytics to the real estate industry.

Building this model will allow AppFolio to analyze the results of our recommender system to assess how interaction data can be used to better their products. We explored multiple models including the Bilateral Variational Autoencoder (BiVae), TriVae, Visual Bayesian Personalized Ranking (VBPR), and Deep Cross Network (DCN) in an effort to achieve the highest prediction accuracy using 16 property features from our given dataset and images. Ultimately, we want to integrate our model into the Santa Barbara property website, which was built by the AppFolio Sponsored CS Capstone team.

California Cooperative Oceanic Fisheries Investigations (CalCOFI)

Interactive Spatio-Temporal Visualization of Long-Term Physical Characteristics of California Coastal Waters

Student team: Prerana Kottapalli (PK), Mallika Gupta, Teresa Gonzalez, and Dante Coletta

Project advisors: Erin Satterthwaite and Trevor Ruiz

The goal of this project is to visualize the variation of ocean chemistry over time and space in the California current to help understand changing ocean conditions and identify potential impacts on marine life. We will create an exploratory tool using long-term oceaongraphic data collected by CalCOFI. Data includes quarterly measurements of oceanic samples, at a depth of up to 1,000m, collected from 1970-2020. Parameters explored are dissolved oxygen, temperature, salinity, and chlorophyll.

Carpe Data

Multi-class Claims Activity Classification Based on HTML data

Student team: Annie Huang, Anum Damani, Rithvik Vobbilisetty, Alex Rudolph, and Tyler Chia

Project advisors: Crystal Zhang, Kevin Neal, and Joshua Bang

Insurance companies typically receive a significant number of claims each year. By automatically flagging and classifying web pages that have information potentially relevant to the claims, our project would help to significantly reduce the amount of manual inspection required for potential cases of insurance fraud. Our project focuses on building machine learning classification models that predict and flag whether or not a web page contains evidence about a fraud claim, and identify the specific type of potentially fraudlent activity that is present.

Our dataset contained 42,485 web pages potentially containing claimant activity information identified through proprietary search methods and manually labeled. We classified each page into one of several classes of potentially fraudulent activity. Ultimately, our best performing model was our SVM classifier with a weighted average precision of 85%.

Cheadle Center for Biodiversity and Ecological Restoration (CCBER)

Big Bee: Hair Recognition and Quantification

Student team: Harper Klauke, Jonathan Wong-Lau, Harleen Kaur, and Nicholas Alexander

Project advisors: Katja Seltmann and Joshua Bang

Bee species are decreasing in both number and diversity. As important pollinators, bees are responsible for creating and maintaining the ecosystems that many animals and humans rely on. Examining their physical traits and the differences between species will allow us to study the resiliency of different bees using computer vision and machine learning. Our objective in this project is to use computer vision techniques to detect and count bee hairs from high-resolution lateral images.

We worked with 550 high resolution lateral bee images that were gathered from various institutions for the Big Bee Library, an online repository of bee image, trait, and specimen data. Contributors included the Museum of Comparative Zoology, Harvard University, University of Kansas Natural History Museum Entomology Division, and the UC Santa Barbara Cheadle Center for Biodiversity and Ecological Restoration. This research was supported by the National Science Foundation: Extending Anthophila research through image and trait digitization (Big-Bee) project (DBI2102006).

Center for Information Technology and Society (CITS)

Analyzing Changes in Social Media Usage Behavior

Student team: Eoin Hayes, Qiru Hu, Jasmine Kwok, Lucas Nguyen, and Xavier Speropoulos

Project advisors: CITS Center, Qing Huang, and Erika McPhillips

In 2020, The Social Dilemma became the most popular movie on Netflix. The documentary reveals the downsides of the major social media platforms. The goal of this project is to analyze the impact of The Social Dilemma on its audience’s attitudes towards social media. Does watching this movie increase/decrease/not change people’s social media usage?

To answer this question, we performed a series of statistical analyses on user-generated posts from Twitter and Reddit. Although our study reveals a difference in the user behavior patterns between the two platforms, the statistical results suggest that, based on the social media data, the movie did not change people’s social media usage.

Energy and Environment Transitions (ENVENT) Lab

Demographic Predictors of Climate Opinion

Student team: Emma Franzblau, Annie Adams, Andrew Bissell, and Johnny Yu

Project Advisors: Matto Mildenberger and Adam Waterbury

The aim of our project is to explore how demographic attributes are differentially predictive of climate opinion in different countries, continents, and time periods. We first cleaned and organized covariate data from 140+ surveys and merged the demographic information into existing data processing routines. Then we developed preliminary and exploratory predictive models to gain insights into our question of interest.

We were ultimately able to set the foundation for further exploration of the data in the next iteration of the project as a result of completing the necessary proof of concepts; which we did by honing in on the demographic of gender as it relates to various climate opinion metrics through different methods of modeling.

See our project poster here.

Earth Research Institute (ERI)

Combining Computer Vision and Satellite Imagery to Detect Center Pivot Irrigation

Student team: Ashley Grinstead, Evan Bui, Tiffany Hsu, and Tristan Chen

Project advisors: Kelly Caylor, Leron Reznikov, and Anna Boser

Irrigated agriculture is expanding rapidly across southern Africa. Because this expansion is occurring without much regulation or monitoring, it’s not exactly clear where and when expansion is happening, or the fate of expanded agriculture. In most expansionary cycles, there is a boom-and-bust dynamic, where many investments fail quickly and only a few persist. The Earth Research Institute (ERI) at UCSB spans areas covering natural hazards, human impacts, earth system science, and earth evolution by collaborating across interdisciplinary departments on campus. Under the guidance of Kelly Caylor (Director of ERI and WAVES lab), we are undertaking a project to explore how center pivot-irrigated agriculture is expanding globally. We’re interested in mapping these dynamics over the recent years and monitoring them going forward.

See our project poster here.

Evidation

Stress Indicators for Frontline COVID-19 Healthcare Workers

Student team: Priscilla Siow, Ali Vakilimafakheri, Sarah Schlund, and Eric Wang

Project advisors: Eric J. Daza and Alex Franks

Using wearable, survey, and biomedical data from healthcare workers measured during the pandemic, we aim to identify variables indicative of stress such as sleep patterns, heart rate, PTSD, etc. We hope that our findings will give insights into how wearable technology can help detect stress in this population.

The features we used primarily came from wearable and survey data such as the Oura Ring, Daily Stress Measure, PTSD, Adverse Childhood Events (ACE), and Demographics datasets that were provided from Synapse database. We built two models: a mixed effects model that accounts for subject-specific baseline stress levels in predicting excess stress; and a logistic regression that identifies population-level indicators of stress.

Wearables Predict Stress in Frontline Healthcare Workers

Student team: Joshua Alvarado, Ian Gascon, Joanne Kim, Mac Kul, and Jessica Nguyen

Project advisors: Julio Vega and Alex Franks

We are using data collected in “The Stress and Recovery in Frontline COVID-19 Health Care Workers Study.” Daily notifications prompted 365 participants to complete tasks through a REDCap study app including a Demographic Survey and a Daily Stress Measure and Work Shift Details survey.

Our project objectives are twofold: identifying the main factors of stress (duration of sleep in different stages, heart rate, respiratory rate, and levels of oxygen saturation); and predicting whether a participant is stressed given those factors.

See our project poster here

HG Insights

Product Recommendation on Scraped Usage Data

Student team: Atherv Gole, Cristian Razo, Eric Cha, Natasha Leodjaja, and Qimin Tao

Project advisors: HG Insights, Rob Fox, and Lauren Wong

The primary goal of this project is to analyze patterns in product adoption to effectively engage with existing and/or potential customers. Our primary dataset consists of companies’ product quality ratings across the United States over a span of 20 years, ranging from 2001 to the present. Two of the main features we used are ‘signal score,’ which is a 1-3 score of how recently the product was observed, and ‘intensity’ which is a measure of how much a product is observed over time.

Neuroscience Research Institute (NRI)

Modeling Neurodegeneration with Single Cell RNA Sequencing

Student team: Jake Stenger, Nancy Zha, and Sara Mandic

Project advisors: Stella Glasauer, Megan Elcheikhali

This project aims to model neurodegeneration in brain organoids by single-cell RNA sequencing. Through the use of statistical methods we are aiming to compare differences in gene expression between control and mutant brain organoids. In addition we aim to identify cell populations that are more or less abundant in the mutant brain organoids. Thus, we are aiming to gain insight in disease mechanisms that lead to neurodegeneration.

Pricewaterhouse Coopers (PwC)

Compression of NLP-Domain Deep Learning Models

Student team: Yan Lashchev, Jorge Murillo, Nathan Roll, Lawrence Su, and Yangyi Zhang

Project advisors: Robert Bernard and Erika McPhillips

The field of model compression has enjoyed many advancements in recent years, yet few reliable methods have been developed specifically for the natural language processing (NLP) domain. In this project, we provide a review of model compression techniques and develop custom NLP-specific compression methods. We demonstrate the implementation of our methods on a sentiment analysis task.

Sprague Lab

Deep Image Reconstruction from Natural Scenes Dataset

Student team: James Du, Michael La, Connor Levenson, and Elise Nguyen

Project advisors: Thomas Sprague and Sikun Lin

We leverage the recent progress made in text-to-image generation for deep image reconstruction. After building a joint-embedding model which maps the patient’s fMRI to a shared image-text space, we use a pre-trained decoder (LAFITE) which is able to recreate the stimulus image from the encoded vector.