Carnegie Mellon University

Research Team Leaders and Projects 

afsaneh

Afsaneh Doryab

Systems Scientist
HCII, CMU

Smartphones and smartwatches have provided the possibility to collect data from the daily life of individuals. In this project we will use passive sensing data from smartphones, smartwatches, and other wearable devices to assess physical and mental health in college students (e.g., depression, loneliness, stress) and behavioral trends (e.g., how stress, sleep, visits to the gym, etc. change in response to college workload -- i.e., assignments, midterms, finals -- as the term progresses). We will explore developing applications to 1) automatically collect physiological and behavioral data from devices, 2) visualize, and 3) analyze the data.

cristina Cifuentes

Cristina Cifuentes
Oracle Labs
Director of Oracle Labs Australia and an Architect at Oracle
http://labs.oracle.com/people/cristina

In the cloud world, the attack surface for an application is much larger than in an on-premise deployment.  Current commonly used languages provide limited support for security, with vulnerabilities such as Injection attacks (e.g., SQL injection, cross-site scripting, etc), and Leaks of sensitive data being the most representative vulnerabilities exploited in web applications over the past 5 years.  Combined with the trend towards micro services, where services are language agnostic and hence are implemented by various languages at a time, we have a growing security problem that needs to be addressed now. 

In this workshop I give an overview of the problem, I’ll review some solutions that have practical applications, and we’ll embark on the first steps to develop a new language that addresses the lack of security concepts provided in today’s commonly used languages for web and cloud applications.  This workshop will rely on the multi-lingual virtual machine GraalVM.  This is a hands-on workshop, software for the workshop will be available prior to the workshop itself.

 

 

carolyn rose

Carolyn Rose
Professor
Language Technologies Institute and Human Computer Interaction Institute
Carnegie Mellon University

Jim McCann

Jim McCann
Assistant Professor
Robotics Institute
Carnegie Mellon University

Lea Albaugh

Lea Albaugh
PhD Student
Human Computer Interaction Institute
Carnegie Mellon University

 

In the modern world, Machine Learning and Automation touch every area of our lives.  In this workshop we explore these topics through the fiber arts.  In fact, one of the first computers was a weaving loom -- the Jacquard loom -- and workshop participants will have the chance to be introduced to a working Jacquard loom as part of the workshop experience!  Like music, many of the underlying paradigms for weaving share the same mathematical foundations as machine learning.  The fiber arts also offer many exciting opportunities for automation.  This workshop will offer both insight into active research on machine learning applied to textile design and automation in weaving as well as hands on experience with smart technologies in a fiber arts group project.

geoff kaufman

Geoff Kaufman
Assistant Professor
Human-Computer Interaction Institute
Carnegie Mellon University

joseph

Joseph Seering
PhD Student
Human-Computer Interaction Institute
Carnegie Mellon University

Conversations about ethical practices in HCI have recently turned a spotlight on the growing prevalence of “dark patterns” - deceptive, covert persuasive strategies that are frequently used by platforms and websites to get users to do things that aren’t necessarily in their best interest: to buy products or upgrades, opt-in to services and correspondences, or remain actively engaged on sites and platforms for longer periods of time, just to name a few.  Because of the subtle and hidden nature of dark patterns, users are generally unaware that they are the target of persuasion, which makes them particularly susceptible to these influences. In this workshop, we will address this basic question: can we design psychologically grounded, easily implementable techniques and strategies to provide users with greater awareness of the dark patterns they’re exposed to - and increase their ability to avoid succumbing to them?

Over the course of the weekend, we will immerse ourselves in the world of dark patterns and the psychological theories and strategies that these persuasive techniques embody - and use this knowledge to inform the development of a prototype of an intervention (or small set of interventions) to empower users toward greater vigilance and resistance to implicit persuasive attempts.  

henny admoni

Henny Admoni
Assistant Professor
Robotics Institute
Carnegie Mellon University

stephanie valencia

Stephanie Valencia
PhD Student
Human Computer Interaction Institute
Carnegie Mellon University

Many people face communication challenges in their daily lives, which may be the result of cognitive disability, motor impairment, or a knowledge gap (as when learning a new language). These challenges often manifest as a slow-down in speech production, which affects pacing and turn taking in conversation. People who have difficulty communicating can face social isolation and a loss of agency if they are not provided enough time to contribute. Social robots may offer a mechanism for supporting people with communication difficulties through conversation regulation mechanisms, such as directing attention to the speaker or moderating turn taking. 

In this project, we will explore the range of behaviors that a simple social robot can use to support conversation between two human partners. Participants will learn about current research in this area and design a set of supportive robot behaviors to increase conversational agency. Initial examples of these behaviors will be implemented on a social robot (Kuri) using Python and ROS. These behaviors will be tested and evaluated within the group. The output of this project will be a set of prototype behaviors for supporting conversational agency for people with communication difficulties.

gabriella marcu

Gabriella Marcu
Assistant Professor
School of Information
University of Michigan

Smartwatches and other wearable devices enable individuals to monitor their activity and mood, helping them to make healthier choices. However, limited research has focused on this type of support for children. This project applies user experience research, human-centered design, and interaction design to understand how smartwatch applications can be designed for children's behavioral self-monitoring.

Laura Dabbish

Laura Dabbish
Associate Professor
Human-Computer Interaction Institute
Carnegie Mellon University

Bogdan Vasilescu

Bogdan Vasilescu
Assistant Professor
Institute for Software Research
Carnegie Mellon University

Huilian Qiu

Huilian Qiu
PhD student
Institute for Software Research
Carnegie Mellon University

 

Representation of women and minorities in technical fields and open-source software development specifically is disappointingly low and getting worse. Open source software development is an important gateway to technical careers and represents a style of open collaboration that is the future of work. Moreover, job opportunities in the field of computing are growing at a faster rate than is possible to fill. Recent work has identified specific barriers to women making initial contributions to open-source projects, and systematic biases by authority holders, e.g., imbalances in pull request acceptance rates. It is relatively unknown, however, how underrepresented newcomers, particularly women, become integrated as full members into open-source communities, and why this happens so rarely. In this project we will analyze participation data from a set of open source projects to examine differences in contribution and communication patterns for male versus female developers. Based on these analyses and literature on the pipeline problem, we will brainstorm about design interventions that could enhance diversity and inclusion in open source communities.

lena

Lena Pons
Machine Learning Research Scientist
CERT, SEI
Carnegie Mellon University

Elli Kanal

Elli Kanal
Technical Manager
CERT Data Science, CERT, SEI
Carnegie Mellon University

 Document classification is a foundational problem in information retrieval. Knowing what content exists in a document collection and being able to retrieve it are two of the most common tasks for organizing and using a text collection. One set of approaches for understanding document contents, broadly referred to as “topic modeling,” identifies sets of words that discriminate documents in different classes. Topic modeling can be used to learn what information a corpus of documents contains, and also to build classifiers to assign new documents to pre-defined classes.

In the workshop we will introduce a basic natural language processing pipeline for document classification. The team will construct a text corpus, investigate various topic modeling techniques, , and select the topic modeling approach most appropriate for their text. Next, the team will identify subjects contained in the corpus based on the topic models and build a classifier. Finally, the team will validate the results of the classifier and determine how the performance can be improved. Participants should expect to learn some basic natural language processing, machine learning and information retrieval concepts.

Lorrie Cranor

Lorrie Cranor
Professor, School of Computer Science and Engineering and Public Policy
Director, CyLab Usable Privacy and Security Lab

Hana Habib

Hana Habib
PhD Student
Societal Computing
Carnegie Mellon University

Josh Tan

Josh Tan
PhD Student
Societal Computing
Carnegie Mellon University

Sarah Pearman photo

Sarah Pearman
Research Staff
Cylab, Carnegie Mellon University

Kyle Crichton

Kyle Crichton
EPP PhD Student
Carnegie Mellon University

Digital voice assistants, like Siri and Alexa, are nearly ubiquitous. They are built into our smartphones and, increasingly, into smart speakers and other devices that we place in our homes. People use them for reasons varying from getting quick updates on the weather to automating their home environment. Devices with voice assistant features enabled, by design, are always listening to their surroundings for "wake words" uttered by their users. Recent news headlines highlight some privacy and security implications related to these smart devices. For example, one Alexa user discovered that her Amazon Echo was sending audio recordings from her home to an acquaintance on her contact list. Additionally, investigators in Arkansas attempted to get access to data from a smart speaker owned by a suspect in a murder case to see if the device captured any important details.  As digital voice assistants become increasingly popular, it is important to better understand how the presence of these devices impacts people’s privacy and security. Questions to explore include: how are people using these devices? Are they aware of the information being collected? How concerned are they about the recordings made by these devices? For this project, we will develop a survey to address some of these questions, and more generally explore user perceptions of digital voice assistants. We will collect and analyze actual user data through Amazon’s Mechanical Turk, allowing us to draw conclusions around these important issues.

norman sadeh

Norman Sadeh
Professor, Co-Director
MSIT-Privacy Engineering, Institute for Software Research
Carnegie Mellon University

Linda Moreci
Lab Manager
Institute for Software Research
Carnegie Mellon University

Abhilasha

Abhilasha Ravichander
PhD Student
Language Technologies Institute
Carnegie Mellon University

Shikun Aerin Zhang

Shikun Aerin Zhang
Language Technologies Institute
Carnegie Mellon University

Privacy is about empowering users to control the collection and use of their data. Research consistently shows that people care about privacy, yet very few take the time to read the text of privacy policies or configure privacy settings. As part of this interactiveworkshop, we will be asking ourselves to what extent we might be able to build computer programs capable of reading the text of privacy policies, and of helping us manage our privacy settings. The workshop will combine short introductory lectures on privacy and on state of the art AI technologies to automatically read the text of privacy policies, learn people's individual privacy preferences and help them configure privacy settings. This will include opportunities to play first hand with AI tools developed by Professor Sadeh and his graduate students, including tools developed to help smartphone users manage mobile app permissions on their phones, tools to help users discover Internet of Things devices around them, technology to automatically read the text of privacy policies and more.

Nathan

Nathan Beckmann
Assistant Professor
Computer Science Department
Carnegie Mellon University

brandon

Brandon Lucia
Assistant Professor
Electrical & Computer Engineering Department
Carnegie Mellon University

The "Internet of Things" (IoT) is going to turn everyday objects into computers. One challenge in building IoT devices is how to deliver energy to them. Batteries are the standard solution, but they are expensive and annoying to replace. Energy-harvesting technology eliminates batteries by letting IoT devices capture energy from their environment instead. Energy-harvesting computers are tricky to program, however, because energy is not continuously available. This project will explore energy-harvesting IoT computing by developing a battery-less electronic game board, where the game pieces are energy-harvesting computing devices and the board itself supplies them with energy. At the end of the project, the team will have built a working board with "smart" pieces that help players play the game.

pallika Pallika Kanani
Principal Member of Technical Staff
Oracle Labs
ed Edward McFowland
Assistant Professor
Information & Decision Science
University of Minnesota
This research project will be an introduction to the burgeoning field of FATE (Fairness, Accountability, Transparency, and Equity) in Machine Learning. With the increasing reliance on machine learning algorithms for recommendations and decision making, their impact on society cannot be overstated. Though the use of machine learning enables many benefits, it is unclear if these benefits are shared equally by all, or worse, if some are adversely impacted. Essentially, there are critical questions that have largely gone unasked: are our algorithms unbiased, fair, and equitable? In this project, we will explore the questions of what does it mean for an algorithm to be biased, fair, or equitable; how do algorithm come to possess these undesirable qualities; and who is responsible for rectifying them? Moreover, we will develop some basic tools to audit algorithm and inspect them for bias. This project will be a combination of discussion and development, so participants should expect to engage in lively conversations around important issues in both society and computer science, as well as build/audit machine learning models on real-world data.  Participants need no previous experience with the ideas of FATE or building machine learning models, but should be interested in both topics and familiar at a high level with basic concepts and goals in machine learning.
nina Maria-Florina Balcan
Associate Professor
Machine Learning Department
Carnegie Mellon University
ellen Ellen Vitercik
PhD Student
Computer Science Department
Carnegie Mellon University

For many computational problems, there are often dozens, hundreds, or sometimes even infinitely-many algorithms to choose from, each of which has varying performance that depends on the specific application domain. For example, clustering problems arise across a diverse array of scientific fields. There, the goal is to group datapoints into subsets based on how similar they are, such as images by subject or proteins by functions. Scientists have proposed an immense variety of clustering algorithms, and the best algorithm for image classification will surely not be optimal for computational biology problems. This workshop explores how scientists can use machine learning to craft algorithms tailored to work extremely well in their specific application domains. We will investigate not only clustering, but also algorithms from computational economics, where we aim to fine-tune selling algorithms to ensure high revenue from a marketplace’s typical daily customers.