Projects 2020


Emnet Mesfin Regassa and Dawit Sefiw Simegn: Hardware-Software Co-Design for Program Analysis
Faculty Mentor: Baris Kasikci

Description: Software analysis techniques such as symbolic execution and model checking rely on exploring the state space of a program for various purposes such as determining the inputs that cause certain program paths to execute, or to find bugs in code. These analyses employ various state space exploration strategies, which ultimately determine their effectiveness and resource usage. In this project we will explore whether existing and novel hardware support could improve state space exploration effectiveness and efficiency.

Simret Araya Gebreegziabher and Aymen Jelaludin Ahmed: Random Forest Based AI Classifier and Explainer

Faculty Mentor: Valeria Bertacco

Description: More and more applications rely on graphs as the underlying data structure: from social networks, to internet's web connections, to brain's neurons and geo maps, and even consumers' product preferences. The performance of these algorithms is often limited by the latency of accessing vertices in memory, whose access present poor spatial locality. The goal of this project is to boost the performance of graph-based algorithms by developing hardware and software solutions to this end: we plan to work on the data layout, on ad-hoc data structures and on designing dedicated hardware acceleration blocks. We hope to boost the performance of graph traversals and computation by 3-5x.

Meron Zerihun Demissie and Kidus Birkayehu Workneh: Privacy-Enhanced Computer Architectures

Faculty Mentor: Todd Austin/Lauren Biernacki

Description: In this project, we will be designing architecture that can process on encrypted data without keys, thereby enabling 100% private computation. We will be designing and testing systems that rival coveted cryptographic techniques like trusted execution environments and homomorphic encryption. Namely, existing processors are equipped with trusted execution environments to preserve data privacy, and homomorphic encryption is seen as the ultimate solution to untrusted cloud environments. But, both techniques have serious drawbacks. Deployed trusted execution environments (e.g., Intel SGX) only encrypt data in main memory. Secret values in the caches or processor are stored as plaintext, allowing exploits like Foreshadow to readily leak private data. While homomorphic encryption never discloses plaintext values, it is currently thousands of times slower than unprotected computations. In this work, we trade trust in hardware for performance. By developing a secure architecture with hardware-enforced private data types, we can assure that secret values are not stored as plaintext and never revealed to software while offering considerable speedups over homomorphic encryption. Through this project, you will work to 1) Add new capabilities to the existing architecture in our simulation platform, 2) Develop real-world privacy-enhanced applications on our system, or 3) Develop techniques to measure the performance and security of the design. These tasks will be assigned based on availability and personal preference.

Gemmechu Mohammed Hassena: Understanding Scenes Using Humans as Rulers

Faculty Mentor: David Fouhey

Description: Our lab is broadly focused on building 3D representations of the world and understanding human/object interaction. Potential projects include learning about: navigating environments, object articulations, commonsense physical properties of objects, and hand grasps. Please look at: for a sense of what projects we’ve done in the past. We will find a specific project based on mutual interest and particular abilities (e.g., stronger systems programming abilities, experience with graphics, etc.).

Simret Araya Gebreegziabher and Aymen Jelaludin Ahmed: Composable Benchmarking

Faculty Mentor: Valeria Bertacco

Description: The 2030 decade will rely on very advanced and specialized computing capabilities, executing several complex algorithms at high performance, so to enable applications ranging from virtual and augmented reality, complex classifications and explainable ML-based decisions, and data analyses at a much more advanced level than it is possible today. The goal of this project is to 1) develop and adapt core algorithms from those future domains to create benchmarks for computing system researchers, 2) design a synthetic, parametric benchmark generator that produces additional benchmarks resembling the same type of workloads -- so to boost the robustness of the offering, and finally 3) develop a library of simple hardware components that implements some of those core algorithms efficiently and can be employed in the development of complete hardware systems.

Tsedeniya Solomon Amare and Bruktawit Teklay Amare: Understanding and Modeling the Effects of Social Behaviors on Virus Transmission to Fight Global Pandemic

Faculty Mentor: Nikola Banovic

Description: Knowledge about the effects of social dynamics on virus transmission and prevention informs development of processes and actions to fight the global COVID-19 pandemic. However, there are many recent policies put in place (e.g., CDC recommendations for transmission prevention to households with family members who contracted COVID-19) without enough evidence to suggest to what extent those policies help stop/slow down the spread of COVID-19. The existing, simple models and simulations have shown potential to explain how different public policies and recommendations impact the spread of COVID-19 and help domain experts persuade policy makers and executive governments to take certain steps. However, to make accurate decisions about policy it may be necessary to go beyond simple abstract models (e.g., dots randomly moving on a screen to illustrate the spread under various physical distancing policies). People's behaviors are much more complex: they are purposeful and triggered by the situations in which their behavior is situated. In this project, we aim to develop detailed computational models of realistic human behavior (in particular mobility) that accurately simulate people's actions in realistic environments to predict the spread of COVID-19. It is our goal to create models that consider people's socio-economic status, political orientation, and exposure to various information about COVID-19 (including misinformation). Our computational models of people's behaviors will enable reasoning and decision-making about different what-if situations to inform current policy to address this global challenge.

Bethel Roger Hall: Correct-By-Construction Neural Networks for Safe Flight Systems

Faculty Mentor: Jean-Baptiste Jeannin

Description: The next generation of aircraft collision avoidance systems use Markov decision problems to optimize alerting logic, then compress the result in a deep neural network representation. However, ensuring that the neural network always issues safe advisories is of critical importance. In this project we will explore ways to ensure that the neural network always issues a safe advisory, either by verifying the learnt network, or by influencing the learning process.


Eden Benti Mesfin: Design and Manufacturing of Large-Scale Inflatable Structures

Faculty Mentor: Evgueni Filipov

Description: Deployable structures that use the principles of origami could lead to applications in multiple scales and disciplines from biomedicine to space exploration. In architecture and civil engineering reconfigurable facades could adapt to the environment, and rapidly deployable shelters and bridges could be used for disaster relief efforts. The objective of this project will be to explore how to scale-up principles of origami for structural engineering applications. The student will first create an analytical model to study the motion and geometry of an origami inspired deployable structure. Next, a laser cutter will be used to fabricate panels for a scaled prototype of the structure. These individual panels will then be interconnected with metallic or plastic hinges that allow for deployment and reconfiguration. The systems will be constructed to minimize the stowed volume, while allowing for a reliable deployment that requires minimum force input. Time permitting, the student will conduct experimental testing to quantify the stiffness of different deployable systems.


Shalom Abebaw Bekele and Biruk Solomon and Samson Hailemichael Woldu : Waste to Energy

Faculty Mentor: Margaret Wooldridge

Description: Around the globe, thousands of tons of food waste are sent to landfills each year. This food waste is a resource that can be used as feedstock for valued products such as fuels and fertilizers. For example, food waste can be converted to biogas for power generation using anaerobic digestion. Necessary first steps in converting food waste to electricity and other products is a quantitative assessment of the resource (i.e., supply) and product needs (i.e, demand). This project involves three critical aspects to assessing and enabling food waste to energy systems:

1. Quantitative assessment of the characteristics of waste in Ethiopia. This includes understand the amount of waste generated (e.g., mass per month), the composition (e.g., plastics, metals, organics (food scraps), and the distribution geographically.

2. Cataloging methods to characterize food waste (and plastics, time permitting) and the outcomes of the analysis (e.g., heating value, elemental composition, moisture content, etc.). Specific attributes for food waste in Ethiopia are the first priority; however, broader scope may be necessary to collect a broad range of data.

3. Cataloging existing sites of waste-to-energy technology demonstrations in Ethiopia, including anaerobic digestion, gasification, and pyrolysis, systems. The performance of the systems will be compared in terms of capacity (i.e., the amount of waste that can be processed per year), efficiency (i.e., waste mass to volume of fuel produced), and scale (type and amount of products such as biogas, biooil and biochar produced annually).


Habtamu Tadesse Abebe and Yohannes Nakachew Zewodie: Simulation Model Development for Electrical Machines and Drives

Faculty Mentor: Heath Hofmann

Description: This project consists of the development of computationally efficient yet accurate Simulink models of electric machines and drives. Nonlinear behavior such as magnetic saturation of the electric machine will be considered. The goal of the models is also to accurately estimate losses and system efficiency of the machine/drive. Potential applications of the developed models are powertrain simulations for electric and hybrid electric vehicles.