Tech + Research Workshop

Most applications use a database system to store state for that application. Most database systems/data stores require advanced knowledge of the structure of the data to be stored for an application in order to be able to compress and access stored data state reliably and quickly. The structure of the data is defined in what’s called a “database schema”. Unfortunately, modern applications are highly dynamic, and the structure of the state for the application changes over time. A schema is changed via a process known as “schema evolution”. Schema evolution is a complex operation that may require restructuring the existing data, which may require moving large amounts of data around. This restructuring process takes time and results in temporary inconsistencies in a data set while it is ongoing. In order to perform these complex changes safely and reliably, application administrators have historically resorted to shutting down the services to perform schema evolution, which results in perceived system downtime by the end user. Unfortunately, this unavailability is unacceptable for many modern applications that require “around the clock availability”, such as telecommunication systems, payment systems, medical facilities, etc. As a result, there has been some recent efforts on developing online schema evolution systems that aim to remove or reduce the downtime during offline schema evolution. In this project, we aim to evaluate these existing online schema evolution tools on a wide variety of schema migration workloads and identify their limitations.

Researchers

Daniel Abadi
Darnell-Kanal Professor of Computer Science UMD Department of Computer Science

Souvik Bhattacherjee
Post-Doctoral Associate UMD Department of Computer Science

This project will investigate opportunities provided by augmented reality (AR) technology to support learning of foreign languages. AR devices allow users to perceive (see and hear) additional information that is blended seamlessly into their physical environment. This provides many possibilities to enhance the environment of a user with information in a foreign language. For example, objects in the user's environment could be tagged with their names in a foreign language and text-to-speech synthesis could be used to pronounce their names. In addition, the user could interact with virtual objects or even avatars in a foreign language, which will be seamlessly presented in the real environment of the user. This project will leverage AR displays and existing services for object recognition, machine translation, and AR rendering to develop a prototype app that explores the possibilities of these technologies for foreign language learning.

Researchers

Matthias Zwicker Reginald Allan Hahne Endowed E-nnovate Professorship UMD Department of Computer Science

Yue Jiang
Graduate Student UMD Department of Computer Science

The Flycycle is a unique virtual reality based system, developed by Advanced Technology and Research Corporation in Beltsville, MD, that integrates both upper and lower body workout through a 2 degrees-of-freedom platform; the upper body motion allows to control the platform’s roll and yaw angles, whereas a cycling system with variable resistance allows for a lower body workout. Oculus Rift is integrated with the Flycycle to achieve a virtual reality system. It is comprised of a head mount display, two touch controllers, and two sensors to translate the user’s head movement into virtual reality. Students will have the chance to explore the Flycycle’s potential as an immersive exercising machine, or even a rehabilitation system, while learning to program games and perform sensor fusion at the cutting edge of the virtual reality world.

Researchers

Axel Krieger
Assistant Professor UMD Department of Mechanical Engineering

Lidia Al-Zogbi
Graduate Student UMD Department of Mechanical Engineering

Property based testing is an increasingly popular approach to software testing where the programmer specifies "properties" that their code should satisfy (for example, applying a list reverse function twice should result in the original input list). The goal of a property based testing tool is to find minimized example inputs that violate the specified properties. Property based testing allows for better input coverage than techniques like unit testing, where the behavior of code is evaluated on a few programmer-selected inputs, and generates more useful inputs than random testing, where the behavior of the code is evaluated on many randomly-generated inputs.

In this project, participants will evaluate the performance of an existing property based testing tool for the Python language. The scope for this project includes comparing property based testing tools to random testing tools, comparing property based testing tools for Python to similar tools for other languages, and finding bugs in real-world Python code.

Researchers

Michael Hicks
Professor, Associate Chair of Undergraduate Education UMD Department of Computer Science

Sankha Narayan Guria
Graduate Student UMD Department of Computer Science

Kesha Hietala
Graduate Student UMD Department of Computer Science

We are increasingly moving to a world where mobility is a shared resource. Bike-sharing, car-sharing, and scooter-sharing are common in most urban places in the United States. While shared mobility has numerous advantages for the end-user, there are numerous issues that cities and the service providers are grappling with. For example, it is not uncommon to see scooters piled up at popular destinations. This project will design innovative solutions for improving shared mobility services. In particular, we encourage you to think of solutions that will leverage the fact that such vehicles can autonomously drive and redistribute themselves. Dataset of bike locations and demands from Capital Bikeshare will be provided.

Researchers

Pratap Tokekar
Assistant Professor UMD Department of Computer Science

Deeksha Dixit
Graduate Student UMD Department of Computer Science

Learning a probability model from data is a fundamental problem in machine learning and statistics. Building on the success of deep learning, there has been a modern approach to this problem using Generative Adversarial Networks (GANs). GANs view to this problem as a game between two sets of functions: a generator whose goal is to generate synthetic samples that are close to the real data training samples and a discriminator whose goal is to distinguish between the real and fake samples. In this project, we aim to leverage some of the state-of-the-art GAN architectures to produce realistic but fake samples from a complex image dataset.

Researchers

Soheil Feizi
Assistant Professor UMD Department of Computer Science

There is a growing interest in technologies designed to foster ‘aging in place’ in today’s world. Imagine growing older in a familiar place, with a variety of technology supporting your lifestyle, health & safety at any and every age! ‘Aging in Place’ technologies range from health-monitoring wearables to motion sensors that remind you of daily activities. Nowadays many researchers and developers are taking an interest in designing technology for people with dementia. While the advancements in this domain are impressive, the acceptance & adoption rates of these technologies remain quite low. This is partly due to the lack of employing a user-centric approach while designing technologies meant particularly for people with dementia. Join us as we immerse ourselves in designing to address these technological challenges! Your design contributions and evaluation ideas will aid important and real research that is currently being conducted by UMD researchers. We are looking for fresh perspectives and creative minds. No prior experience with design or technology is required!

Researchers

Amanda Lazar
Assistant Professor UMD College of Information Studies

Kausalya Ganesh
Graduate Student

Shaan Chopra
Graduate Student

Emma Elizabeth Dixon
Graduate Student

Hanuma Teja Maddali
Graduate Student

This project will develop a Virtual Try-on App that can estimate a 3D model of a human body and its outfit directly from a few photographs with little human interaction. Our current system can capture the global shape and geometry of the clothing, and extract the physical and geometric properties of cloth. Unlike previous methods using full 3D information (i.e. depth, multi-view images, or 3D scanned geometry), our approach can achieve garment recovery from a set of single-view images by using physical, statistical, and geometric priors and physics-based cloth simulation for Virtual Try-on of customized apparel to improve the user's online shopping experience.

Researchers

Ming Lin
Chair, Professor, CRA DREU Co-Director UMD Department of Computer Science

Yu Shen
Graduate Student

Calling contexts, or call paths, refer to the sequence of function invocations that lead to a statement in the code. A calling context tree (CCT) is a prefix tree of all call paths in a program. CCTs allow analysts to understand differences in function behavior that depend on how/from where the function was called. Calling context trees of parallel programs couple performance metrics with call paths, helping understand their execution and performance.

To identify performance bottlenecks, programmers and performance analysts visually explore CCTs to form and validate hypotheses regarding degraded performance. However, due to the complexity of parallel programs, existing visual representations do not scale to large programs running on a large number of processors. The goal of this project is to explore different visual analytics techniques to simplify the visualization of large CCTs.

Researchers

Abhinav Bhatele
Assistant Professor UMD Department of Computer Science