AURA 2023 Projects

Aklilu Zenebe Dejene    

Presentation: "Integrating NaSICON Membrane into Redox Flow Battery"

Project Title: Integrating ceramic membranes into redox-flow batteries   

Faculty Mentor: David Kwabi       

MECHANICAL ENGINEERING   

Description: Aqueous redox-flow batteries are a promising technology for grid-scale storage of renewable (e.g. solar and wind) energy. Integrating solid ceramic membranes into these batteries can enable inexpensive charge-storing materials and thus low system costs, which will boost their practical viability. However, there are several unanswered scientific questions about the chemical, electrochemical and mechanical stability of these membranes during operation in flow cells. The student involved in this project will contribute to answering these questions by assembling flow cells with ceramic membranes, and testing their performance under a wide variety of cycling conditions.

Yared Tegegn Gebeyaw

Presentation: "A Carbon-aware Scheduler minimizing the Carbon Emission of Computing"

Project Title: Accelerating Explainable AI Inference  

Faculty Mentor: Valeria Bertacco

COMPUTER SCIENCE AND ENGINEERING    

Description: Machine learning algorithms are being adopted in a growing number of applications. Some of these applications have critical impacts in social and human aspects, thus there is a pressing need for AI predictions to be explainable, that is, justifiable to end-users. Some algorithms are more amenable to explainability, thus adhering to new legislation, and leading to new task-specific insights. However, their execution through software applications running on multicore processors does not provide results at the pace needed by their users: for some applications (e.g., medical diagnoses), models may take hours, days, or weeks to train, while for others (e.g., autonomous vehicles) decisions are required within milli-seconds. Such demands lead to the need for specialized hardware solutions in this space. This project focuses on the development of specialized architectures and/or memory hierarchies to address this need. 

Christina Solomon Hailu

Presentation: "AI-Enhanced Medical Training: Closed-loop Communication Annotation Tool"

Project Title: Multimodal Virtual-Reality System in Simulation-based Emergency Medicine Training

Faculty Mentor: Alanson Sample

COMPUTER SCIENCE AND ENGINEERING

Description: Existing healthcare training for cardiac arrest focuses on real-world manikin-based simulations where instructors give direct feedback on student technique, team dynamics, and quality of care. However, this type of training is highly dependent on the students having access to state-of-the-art training facilities and can lead to inconsistencies in feedback from instructor to instructor. 

In order to generate an effective and scalable training method, this project leverages a newly developed multi-participant, Virtual-Reality cardiac simulation tool developed by the UM medical school, allowing for anyone with a VR headset to participate in training from anywhere in the world. This SURE project aims to develop an analytic system to evaluate learners' cognitive (e.g., clinical decision-making) and behavioral (e.g., situational awareness, communication) processes using data from VR simulations along with an array of on-body sensors. SURE students will work with a team of CSE graduate students and faculty, along with clinicians and educators from the UM medical school to develop, deploy, and test this VR training assessment tool.

Ephrem Alemayehu Lemma       

Presentation: "Smart manufacturing and industry 4.0"

Project Title: Smart manufacturing and industry 4.0        

Faculty Mentor: Dawn Tilbury

ROBOTICS             

Description: Computing and networking technologies are becoming pervasive in manufacturing systems, enabling vast amounts of data coming from the factory floor to be used to improve productivity and quality, thereby reducing costs for consumers. There are many open research questions on how to best leverage this data, to build models of the system operation, predict future outcomes, and adapt the system to disruptions.  The summer student(s) will work in a lab with both additive (3D printing) and subtractive (CNC machining) processes, several collaborative robots, and high-performance simulations.  Data from the machines and robots will be used to build models that can be encapsulated in "digital twins" which can improve the overall system operations.

Rediet Ferew Teka   

Presentation: "SYNGEN: A Synthetic ML Dataset Generator and Expander"

Project Title: Sustainable Computing Metrics and Solutions

Faculty Mentor: Valeria Bertacco       

COMPUTER SCIENCE AND ENGINEERING

Description: Computing is a significant source of energy and environmental overheads worldwide. In 2022, the carbon emissions from Information and Computing accounts for 3% of worldwide carbon emissions, on par with those from the aviation industry. The trend is exacerbating, with a projection that the worldwide carbon emissions will reach 8% in the next decade. There is a pressing need to design sustainable green applications with minimal carbon footprints. While there is initial research in measuring, evaluating and limiting the emissions associated with specific applications, or families of applications, this is a nascent field leveraging very preliminary estimates. This project will focus on improving current emissions estimation frameworks and proposing a user-facing tool to suggest actions that can improve on an overall computation’s emissions.

Beimnet Bekele Guta   

Presentation: "Robot Path Planning with Continual Learning"

Project Title: Machine Learning for Robot Motion Planning    

Faculty Mentor: Dmitry Berenson

ROBOTICS

Description: This project focuses on exploring machine learning methods for use in robot motion planning. The project will begin by implementing and testing existing baseline algorithms for learning dynamics models and constraints for use by a motion planner. Then we will explore how to build and improve on the state-of-the-art methods to enable faster and more accurate planning of robot motion. Example applications will include manipulating deformable objects such as cloth and rope.

Gemechis Urgessa Guyo

Presentation: "Synchronous Programming and Verification with Refinement Types"

Project Title: MARVeLus: A Robotics Platform for Verification and Implementation

Faculty Mentor: Jean-Baptiste Jeannin 

ROBOTICS

Description: Robots and other embedded systems often find themselves in safety-critical applications, where reliability is paramount. Failures can be costly or dangerous, and testing can be inconclusive, so formal verification is required for any rigorous guarantees of safety. MARVeLus, developed in our lab, is a robotics platform designed with formal verification of real-life systems at the forefront. Typically, languages in this space either enable verification but do not produce directly executable code, or are not easily verifiable. MARVeLus aims to enable both verification and execution, giving the user the assurance that what they verify is what gets executed on the actual system. This is accomplished by combining the stream-based nature of synchronous programming with the compile-time verification afforded by type-checking. MARVeLus comprises the type theory, verifier, runtime, and robot drivers that put this theory into practice. Currently, MARVeLus allows verification of certain temporal properties and deployment to small wheeled robots. We plan to expand the scope of MARVeLus over the summer, by allowing the user to specify richer safety properties and bringing more robotics platforms into the MARVeLus ecosystem. Experience in one or more of robotics programming, embedded systems programming, type theory, or functional programming is desired.

Tesfaye Adugna Hordofa

Presentation: "Formal Verification of Processor Robustness against Silent Data Corruptions"

Project Title: Formal Verification of Processor Robustness against Silent Data Corruptions      

Faculty Mentor: Yatin Manerkar

COMPUTER SCIENCE AND ENGINEERING      

Description: Silent Data Corruptions (SDCs) consist of physical-layer faults in microprocessors which do not result in the processor failing. Instead, the failures are "silent"; the only indication that they have occurred is that the processor's computation generates incorrect results, often in a repeatable fashion. Furthermore, SDCs can manifest in processors long after customers start using them, so they cannot be detected at fabrication time. Recently, the percentage of processors that exhibit SDCs has risen significantly, causing problems for companies that use large quantities of processors, such as cloud providers. Major industry players delivered a "call to action" for research to help detect and mitigate SDCs in a panel at ISCA 2022.

This project aims to help prevent hardware faults from manifesting as SDCs by formally modelling processors and verifying whether any occurrence of a given fault pattern (e.g., a single bit flip, multiple simultaneous bit flips, etc) could result in an SDC. If a potential SDC is detected, a hardware engineer could protect against it through mechanisms such as duplicating computation and comparing the results, and then re-run the verification. The end goal of the project is to be able to verify processors against SDCs caused by certain fault patterns.

Hana Andargie Kassie

Presentation: "Human-autonomous Vehicle Interaction and Teaming"

Project Title: Autonomous Vehicles, Trust, and Situational Awareness 

Faculty Mentor: Dawn Tilbury   

ROBOTICS 

Description: The MAVRIC lab at the University of Michigan is conducting research on the subject of human-autonomous vehicle interaction and teaming. This research is highly interdisciplinary and current projects have broad scopes including improving performance of teams of humans and robots, increasing and managing trust, and optimizing situational awareness during complex scenarios and dynamic missions. For this summer project we plan to spin up a new research idea stemming from recent work on mental-models and human-robot teaming with multiple autonomous vehicles. This project allows undergraduate researchers the opportunity to come in at the ground floor of the research process and work with us to build a research question, decide on an approach to answering it and with luck conduct early pilot testing. 

Petros Beyene Mola

Presentation: "Leveraging Loop Optimizations in Data Oblivious Algorithms"

Project Title: Microarchitectural Side Channel Discovery via Machine Learning    

Faculty Mentor: Todd Austin

COMPUTER SCIENCE AND ENGINEERING     

Description: Side channel attacks are one of the primary means to exploit modern hardware. These attacks use visible properties of a system, like timing, power, and electromagnetics, to infer information about the secrets held within. Eliminating these vulnerabilities is a primary focus of hardware security research today. In this work, we are developing a tool to help computer architects locate side channel leakages in their designs, using advanced machine learning techniques. The tool collects various observable properties of a system under test and applies statistical analyses and machine learning to evaluate the system’s security. A key goal of the project is to find new side channels using advanced AI in an explainable manner.

Eyerusalem Abate Tegegn   

Presentation: "Introducing Axiomatic Concurrency into the HardKAT Compiler"

Project Title: Privacy-enhanced computer architectures     

Faculty Mentor: Todd Austin 

COMPUTER SCIENCE AND ENGINEERING       

Description: In the age of big data, privacy is a key concern in sharing data. Unfortunately, the computing world is riddled with stories of security attacks… even for the most secure enclaves. The solution we want to investigate with this project uses encryption technology to encrypt data locally, transfer it to the cloud for any required computation, and receive encrypted results back. The enhanced cloud system performs the computation directly on the encrypted data without an access key -- it never accesses the plaintext data nor can it decrypt the sensitive data. Only the end device can decrypt the result and store it locally. This project will focus on the design and evaluation of computer architectures that can execute directly on encrypted data in ways that are both confidential and tamper-proof, thereby removing any opportunity for attackers to see or manipulate sensitive data.