Clinic Projects

Please click on a link below to view the Harvey Mudd College Computer Science Clinic projects for the corresponding time period.

Clinic Projects for 2017-2018

Genetic Gameplay: Showcasing 23andMe's Personal-Genetics API

Client
23andMe

Faculty Advisor
Julie Medero

Student Team
Teal Stannard (PM), Giovani Barrios-Arciga, Elise Cassella, Jacob Rosalsky
23andMe is a bioinformatics company whose mission is to help people access, understand, and benefit from the human genome. 23andMe analyzes DNA to identify specific variations, or alleles, in order to determine ancestry and wellness traits. 23andMe makes much of this information available to third party developers through a public API. The goal of this project is to create an application which will help encourage other developers to use the public API. Therefore, the application should serve as an example of how to correctly access the public API and securely handle a user's genetic information. Additionally, we aim to engage users while their data is being analyzed.

3D Model Integration for Agile Digital Product Development

Client
Accenture Labs

Faculty Advisor
Ran Libeskind-Hadas

Student Team
Emily Dorsey (PM), Wenbo (Tracy) Cao, Holly Mitchell, Shailee Samar, Abigail Schantz
3D asset management can pose a significant problem for Augmented and Virtual Reality (AR/VR) applications. The Harvey Mudd Clinic Team is working with Accenture Labs to make AR/VR content management easier by providing a straightforward workflow allowing users to alter their content from within existing tools. The team's Unity plugin allows inexperienced users to understand the tradeoffs between the visual quality of their assets and app performance, and then manage that tradeoff by customizing their asset's polygon counts.

Amazon Music Graph

Client
Amazon Music

Faculty Advisor
Robert Keller

Student Team
Dan Diemer, Rebekah Justice, Renata Paramastri, Amelia Sheppard (PM), Cha Suaysom
Our project focuses on creating an enhanced infrastructure to enable Amazon's intelligent personal assistant, Alexa, to respond to a broader and deeper range of queries about music. To this end, we developed an ontology (graphical knowledge structure) for recorded music and prototyped a database-backed, text-based web interface based on the ontology, which demonstrates responses to sample queries.

Image-Text Classification to Correct the Amazon PrimeNow Search Experience

Client
Amazon Prime Now

Faculty Advisor
Yekaterina Kharitonova

Student Team
Alex Mitchell, (PM-S), Zhepei Wang (PM-F), Kofi Sekyi-Appiah, Tina Zhu
Amazon Prime Now wants to automate the process of identifying products whose images do not match their associated titles on their website. Our team has designed and implemented a solution using various machine learning techniques to automatically identify these problematic listings.

Traffic Testing in Fully Programmable, 6.4Tbps Networks

Client
Barefoot Networks

Faculty Advisor
Geoff Kuenning

Student Team
E. Taylor Yates (PM), Gus Callaway, M Sangheetha Naidu, Matthew Gee
Barefoot Networks builds high-speed programmable networking devices. Our team is building a configurable network traffic tester that can handle up to 6.4Tbps of traffic—enough for every student at the Claremont Colleges to simultaneously stream 200 high-definition movies. The tester will send varying traffic patterns to the device under test, which will immediately forward the packets back to the tester. The tester will analyze packet delay, sequencing, and loss, and will gather and report relevant statistics.

Algorithmic Fairness: Using Machine Learning To Detect Disparate Impact In Geo-based Risk Modeling

Client
Consensus Corporation

Faculty Advisor
Yi-Chieh (Jessica) Wu

Student Team
Daniel King (PM-S), Abby Tisdale (PM-F), Tiffany Fong, Kyra Yee
Consensus Corporation makes point-of-activation software for the sale of mobile phones. This software includes a risk prediction engine, which predicts whether or not a phone plan will be deactivated and the sale deemed fraudulent. The Consensus Clinic Team's project is twofold: (1) improve this risk prediction engine by incorporating new features and analyzing different machine learning models and (2) detect unintended bias in the risk prediction engine.

Temporal Segmentation of Surgical Suturing

Client
Intuitive Surgical

Faculty Advisor
Colleen Lewis

Student Team
Juliet Forman, Varsha Kishore (PM), Jane Wu, Hyobin You, Angela Zhao
Intuitive Surgical specializes in minimally invasive robot-assisted surgery and they developed the da Vinci surgical system. The objective of our project is to analyze the motion of surgical instruments controlled by surgeons and video recording of surgeries by developing computational models that automatically recognize suturing activities during robot-assisted surgery. Our approach comprises machine learning and neural network architectures. These computational models could then be used to generate advanced analytics such as surgeon performance reports.

Integrating Distributed-Memory Machine Learning into Large-Scale HPC Simulations

Client
Lawrence Livermore National Laboratory

Faculty Advisor
Christopher Stone

Student Team
Amy Huang (PM), Evan Chrisinger, Jeb Bearer, Katelyn Barnes
Supercomputers provide the computing power for complex physics simulations, but these simulations require frequent manual adjustments to prevent run-time failures. Machine learning is a potential solution for automating this process. The LLNL clinic team is developing a machine learning model appropriate for supercomputers that can learn from the output of physics simulations as they run in real time.

Augmented and Mixed Reality for the Driver

Client
Mercedes-Benz Research & Development North America

Faculty Advisor
Geoff Kuenning

Student Team
Aman Raghuvanshi (PM-S), Julio Medina (PM-F), Drew Summy, Meredith Simpson
The objective of this project is to explore the possibility of using augmented reality (AR) while driving a car. This entails comparing various AR headsets such as the Microsoft Hololens and the Meta 2, and building a functioning prototype that addresses a moving car use-case. The headset comparison involves testing devices in a stationary and moving vehicle, and understanding their respective development experiences. Our prototype application attempts to help drivers park more safely and precisely, and serves as a starting point for the Mercedes-Benz team for an extendable AR headset prototyping framework.

Minimizing Communication in Uncertain Multi-Agent Schedules

Client
NASA AMES Research Center

Faculty Advisor
Jim Boerkoel

Student Team
Grace Diehl (PM), David Chu, Marina Knittel, Judy Lin, William Lloyd
Our project aims to provide NASA with new scheduling protocols for multi-agent systems (e.g., a robot team). In such systems, scheduling disturbances (due to environmental factors or equipment malfunction) may necessitate rescheduling. However, communicating new schedules to all agents can be resource intensive. Reducing rescheduling can improve the performance of multi-agent systems when communication resources are limited. Our team has designed three algorithms to minimize rescheduling, trading communication for schedule quality. Further, we developed infrastructure for testing such algorithms.

Opening the Web Analytics Black Box with Innovations in Machine Learning, Visualization, Animation, and Big Data

Client
New Relic, Inc.

Faculty Advisor
Melissa O’Neill

Student Team
Pratyush Kapur (PM-S), Lee Norgaard (PM-F), Alexandre Trudeau, Edward Carroll
Have you ever wondered how your website is performing? Wouldn't it be convenient to know if your website is going to experience problems before they happen? The New Relic Clinic Team is exploring new ways to visualize website analytics data, communicate application health, and predict spikes in errors. Using animation, machine learning, and deep learning, the New Relic Clinic Team is developing innovative and unique solutions for New Relic's customers.

Detecting Phishing Using Deep Learning Networks

Client
Proofpoint, Inc.

Faculty Advisor
Zach Dodds

Student Team
Daniel Sonner (PM), Nic Trieu, Srinidhi Srinivasan, Amberlee Baugus, Montana Roberts
Proofpoint is a leader at identifying online threats. This project seeks to more accurately classify phishing webpages based on their look, i.e., their visual rendering within a browser. The team designed, piloted, and tested software tools to support this investigation, culminating in a machine-learning pipeline that estimates the probability that a page is phishing based on its screen capture.

Securing Today's Software-Development Pipelines

Client
Rapid7, Inc.

Faculty Advisor
Elizabeth Sweedyk

Student Team
Zhenghan Zhang, Spencer Michaels, Eric Nguyen, Sarah Sedky
Our project aims to expand the capabilities of Rapid7's security assessment platform by integrating the company's newly-developed container assessment service into popular continuous integration tools, namely Jenkins, Bamboo, and Teamcity. Anyone developing a container with these tools can add our plugin to the build pipeline to check for vulnerabilities during each build. The plugin generates a detailed assessment report, and the user can configure rules to pass or fail the build depending on various criteria present in the assessment results.

Wood Veneer Classification and Cataloguing, Mobile

Client
Steelcase, Inc.

Faculty Advisor
Katherine Breeden

Student Team
Brenda Castro (PM), Dalton Varney, Jessica Wang, Samantha Andow
Steelcase is looking to ensure color consistency before large purchase orders for their wood veneer furniture. In order to reduce waste due to veneer color issues, our clinic team is developing a portable veneer classification device that can verify veneer colors objectively in uncertain lighting conditions. The prototype is a mobile lightbox with an intuitive graphical user interface and a computer vision classification system.

Machine Learning on DNS Data to Discover Security Threats

Client
Webroot, Inc.

Faculty Advisor
Lisa Kaczmarczyk

Student Team
Julia McCarthy (PM), Anthony Romm, Reiko Tojo, Danny Wang
In this project, the goals were to aggregate DNS-level data, apply machine learning approaches to identify command and control (C&C) botnets through automated analysis of live traffic patterns, and construct a website for dynamic visualization of threats. Visualizations help the user pinpoint where botnet attacks are coming from, identify geographic hotspots for botnet activity, and find out who is at risk for infection.