Master’s Student Research Opportunities

Several Faculty members in the Department of Electrical and Computer Engineering offer short-term or ongoing research projects in which Master’s students may participate for academic credit or for pay.

Descriptions of available projects are listed below, with application deadlines and links for applying.

DetailS

How to Apply

If you are interested in an opportunity, complete the corresponding application. Your application will be forwarded to the project supervisor. The project supervisor will contact you directly if interested in interviewing you.

Updates

Projects are updated at the beginning of each semester and as new projects become available.

Important: Hours Limit

Please remember that all graduate students may earn pay up to 19.9 hours per week only. Additional work authorization is not required for international students for on-campus employment.

Current Research Opportunities

3D Printed Mold Creation for Epoxy Deep Ocean Electronic Devices
Faculty Advisor: Dr. Martin Brooke

We have been experimenting with ways to quickly create arbitrarily shaped epoxy housings for electronic components such as raspberry pi based sensors.  3D printed molds tend to melt, and leak, and the epoxy remains bonded to the 3D printed plastics, even when using mold release.  We have had some promising results with using 3D Printing for Two-Part Silicone Molds. 

Deadline: Open until filled

Compensation Options: MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Application link

3D Printing Biomolecules in the Protein Database
Faculty Advisor: Dr. Martin Brooke

We would like to be able to routinely print biomolecules such as ATP Synthase and ligand gated ion channels from their protein data base files.  This will involve developing a software pathway from the pdb or cif file to stl file for 3D printing. Issues such as how to support the molecules when printing, what style of print to use (e.g., space filling atomic style) will need to be dealt with.

Required Skills: Programming / 3D printing

Deadline: Open until filled

Compensation Options: MS Project Exam, Independent Study

Expected Duration: Two semesters

Application link

Deep Neural Network Game Player: Learning By Example
Faculty Advisor: Dr. Martin Brooke

We will use AI techniques such as Deep Neural Networks to attempt to train software to play online games.  The approach is to create a large database of game play that can be used to train progressively better software players, leading to software reinforcement learning, where the software creates its own training sets to improve itself.

Required Skills: MATLAB, cloud computing

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: Two semesters

Application link

Neural Network Prediction of Ocean pH from Metal Potentials
Faculty Advisor: Dr. Martin Brooke

We will use a Neural Network or other learning software to predict ocean pH from the measured electrical potential of metals in seawater.  There is an opportunity to visit the Duke Marine Lab for testing of the sensors.

Required Skills: Arduino Code, MATLAB

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: Two semesters

Application link

Predicting Strength of 3D Printed Parts for Lightweight Drone Frames
Faculty Advisor: Dr. Martin Brooke

We will use COMSOL / SOLIDWORKS or other software to try to predict the strength of 3D printed parts for use in drone airframes and other carbon fiber frames.  The directional (anisotropic) properties of 3D printed materials printed on low cost printers are of particular interest.

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: Two semesters

Application link

Virtual Reality Interface for MATLAB Visualization
Faculty Advisor: Dr. Martin Brooke

We would like to have a way to visualize surfaces and other 3D objects in MATLAB using a virtual reality headset such as the oculus rift, HTC Vive, or Fove.

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: Two semesters

Application link

Optimizing Cochlear Implant Sound Processor Configurations via Neural Response Properties to Improve Speech Comprehension
Faculty Advisor: Dr. Leslie Collins

Speech understanding by cochlear implant (CI) users is partially governed by the CI processor parameters, and these are typically adjusted by an audiologist in the clinic.  However, the sheer size of the processor parameter space makes it impossible to test every combination of parameter values manually in the clinic to optimize speech understanding. Dr. Collins’ lab in collaboration with Dr. Tobias Overath (DIBS), Dr. Mike Murias (DCABD), and Dr. Josh Stohl (MedEl) have recently received a DIBS Incubator Award for a project that investigates whether EEG data recorded while a CI subject is listening to speech might be used to automatically optimize CI processor parameters. We are looking for an MS student to help us develop estimation and optimization algorithms to investigate our hypothesis that CIs can be automatically tuned using EEG data.  Optimal candidates for this position would have a background in signal processing, detection and estimation, and optimization, and would have programming experience in Matlab or Python.

Required Coursework: Random Signals and Noise (required), Signal Detection and Extraction Theory (preferred)

Required Skills: Signal Processing, Machine Learning, Matlab, Python

Deadline: Open until filled

Compensation Options: MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Application link 

Modeling and Testing Autonomous Forms of Transportation
Faculty Advisor: Dr. Missy Cummings

Interested in autonomous forms of transportation like driverless trains, planes and cars? The Duke Humans and Autonomy Laboratory (HAL) is looking for researchers interesting in developing simulation models of dispatch centers for all types of autonomous vehicles. In addition, HAL is also looking for students interesting in developing a motion capture environment for small drones. 

Required Skills: Java/C/C++

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study, Paid ($20/hr)

Expected Duration: Can be scoped as needed

Application link

Deep Learning Based Retinal Image Analysis
Faculty Advisor: Dr. Sina Farsiu

Students interested in automated medical image analysis of ophthalmic images using deep learning are invited to participate in this multidisciplinary project.  Students should have a basic knowledge of image processing.  Examples of previous similar projects in PI's lab can be found here: http://people.duke.edu/~sf59/

Required Coursework: Digital Image Processing

Required Skills: Matlab (Required), Python programming skills (Preferred)

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study, Paid ($15/hr)

Expected Duration: One semester with potential to extend

Application link

Radiation Survivability of MEMS Microelectronic Circuits with Carbon Nanotube Field Emitters
Faculty Advisor: Dr. Jeff Glass; Resarch Mentor: Dr. Jason Amsden

Solid-state technology dominates the consumer electronics market because of the low cost associated with large-scale integration. However, there are numerous applications in which solid-state devices are unreliable or do not provide adequate performance, particularly applications with military systems operating in high radiation environments. In these applications, vacuum microelectronic devices present an attractive alternative. Despite the performance advantages, the use of vacuum electronics has been limited because there is no versatile and reliable microscale platform that enables integration of large numbers of vacuum circuit elements on a single substrate. To address this need, RTI International in collaboration with Duke University has been developing a Microelectromechanical systems (MEMS) platform that enables integration of high-performance microelectronic vacuum components into functional circuits on a single silicon substrate. A variety of devices, including vacuum triodes and ion sources, have been demonstrated using this platform which combines versatile and well-established polysilicon fabrication technology and integrated carbon nanostructure cold-cathode field emitters. While these devices avoid the radiation-induced charge carrier problems in solid-state devices, other effects of radiation on this MEMS platform have not been studied. The objective of this study is to determine the effects of radiation on this device platform in collaboration with the NASA JPL Radiation Effects Group and the Naval Research Laboratory.

Required Coursework: Introduction to Micro-Electromechanical Systems, Electromagnetic Theory

Required Skills: Hands-on experience with vacuum systems and materials characterization (SEM, Raman spectroscopy)

Deadline: September 14, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Paid ($16/hr)

Expected Duration: One semester with potential to extend

Application link

In-Situ Functionalization of Carbon-Based Materials for Electrochemical Liquid Disinfection
Faculty Advisor: Dr. Jeff Glass; Research Mentor: Dr. Edgard Ngaboyamahina

The primary routes for human exposure to microbial pathogens from wastewater arise in agricultural settings. Good results for pathogen removal or inactivation have been obtained by membrane filtration, UV irradiation, and pasteurization. However, such treatments have high investment costs and require maintenance that may not be feasible for developing countries. Electrochemical disinfection presents an alternative technology that is effective, simple to operate, low cost, and energy efficient. Some key benefits provided by this method include in situ generation of disinfectants, no storage of chemicals, and a straightforward low maintenance process. This is especially the case for blackwater treatment, as the presence of urine provides sufficient electrical conductivity for electrochemical processes to be conducted without additional chemicals
Our previous work investigated the surface chemistry of a boron doped ultrananocrystalline diamond electrode and its role in efficiently producing a powerful disinfectant, namely H2O2. The project aims at improving the use of oxidants in an electrochemical disinfection system for the treatment of liquid human waste. The main goals are: (i) to reduce the energy needed for electrochemical liquid disinfection (ii) to transfer the fundamental understanding gained on diamond to cheaper carbon materials-based electrodes such as graphite and (iii) to assess the performance of an electrochemical cell using 3D electrodes.

Preferred Coursework: Foundations of Nanoscale Science and Technology and/or Introduction to Solid-State Physics

Preferred Skills: Hands-on experience, materials characterization (e.g. SEM, XPS and XRD)

Deadline: September 14, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Paid ($16/hr)

Expected Duration: One semester with potential to extend

Application link

Odor Mitigation for On-Site Sanitation Technologies
Faculty Advisor: Dr. Jeff Glass; Research Mentor: Dr. Edgard Ngaboyamahina

Fifteen percent of the global population practice open defecation. Preventable diseases such as diarrhea linked to open defecation and poor fecal sludge management are among the highest causes of illness and death, especially of children, in developing countries. Feces defecated in the open come back to humans through many ways, as fecal-oral transmission routes. In those regions, decentralized waste treatment systems are one solution to the problem of open defecation that creates vulnerability, particularly for women and children who are exposed to a loss of dignity, abuse, or harassment while defecating in the open. However, malodor is a major risk factor in the adoption of effective sanitation technologies, as foul-smelling sanitation facilities cause people to practice open defecation.

Most common odorant molecules consist of a carbon backbone ending with functional groups such as aldehydes, alcohols, or ketones. Interestingly, the change of the chemical functional group on a common carbon backbone can result in dramatically different odorant perceptions, from fruity to waxy or grassy for instance. The study proposes to demonstrate the capability to modulate malodor and generate a pleasant olfactory perception simply by applying an electrical potential to the offending liquid source to induce an electrochemical reaction. The project includes the following steps: (i) Determination of the theoretical oxidation-reduction (redox) potentials of a set of malodorant molecules by quantum mechanics calculations, coupled with electrochemical experimental validation using carbon-based electrodes, (ii) Verification of the odorant molecules’ ability to switch from one smell to another by real time sensory and chemical analysis, and (iii) Construction of a flow-through odor modification device.

Preferred Coursework: Nanoscale and Molecular Scale Computing

Preferred Skills: Nanoscale and Molecular Scale Computing

Deadline: September 14, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Paid ($16)

Expected Duration: One semester with potential to extend

Application link

Numerical Modeling of a 3D-Electrochemical Cell for On-Site Sanitation Technologies
Faculty Advisor: Dr. Jeff Glass; Research Mentor: Dr. Edgard Ngaboyamahina

Across the world, it is estimated that 2.7 billion people do not have access to adequate sanitation. Poor sanitation causes severe health risks, which kills 1.5 million children each year. Yet, there is still typically no treatment system in place to deal with the resulting wastewater, efficiently and in an affordable way. Established disinfection techniques like chlorination and ozonation require storage of chemicals and generate harmful byproducts, which present a major drawback to small-scale treatment plant safety. Electrochemical disinfection presents an alternative technology that is effective, simple to operate, low cost, and energy efficient. Some key benefits provided by this method include in situ generation of disinfectants, no storage of chemicals, and a straightforward low maintenance process.

The project will develop a theoretical comparison of a packed bed electrochemical reactor vs other geometries using finite element software (Comsol). The potential exists for a packed bed reactor to be more energy efficient with a higher disinfection rate than a standard parallel plate reactor. This will be determined through this theoretical comparison to determine the distribution of oxidative species and thus the expected efficacy and efficiency of disinfection, taking into account critical parameters such as current density, conductivity, flow rate, and temperature. The optimal parameters obtained during the theoretical study will be applied to a real packed bed for experimental validation.

Preferred Coursework: Probability for Electrical and Computer Engineers

Required Skills: Numerical simulation

Deadline: September 14, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Paid ($16/hr)

Expected Duration: One semester with potential to extend

Application link

Electrochemical Conversion of Carbon Dioxide to Liquid Fuel
Faculty Advisor: Dr. Jeff Glass; Resarch Mentor: Dr. Matt Kirley

An emissions-free energy system is necessary to address the crisis of global climate change. Recycling atmospheric carbon dioxide into chemical fuels would allow more widespread use of renewable energy resources, and using these fuels would result in net-zero emissions. To enable such a system, the Nanomaterials and Thin Films Laboratory is currently developing technology for electrochemical reduction of carbon dioxide to liquid fuels. The student should have experience with electrochemical (electrolysis, voltammetry, and electrochemical impedance spectroscopy) and physical-chemical (chromatography and nuclear magnetic resonance spectroscopy) techniques for synthesis and characterization. The student should have knowledge in fundamental and experimental analytical chemistry and strong laboratory skills.

Required Skills/Coursework: 

Deadline: September 8, 2017

Compensation Options: MS Thesis Exam

Expected Duration: Two semesters

Application link

Circuit and Control of Novel Reconfigurable Battery
Faculty Advisor: Dr. Stefan Goetz

Conventional batteries are hard-wired packs made of individual cells. In all aspects, such as ageing, capacitance, heating, and peak power, the performance of the battery is limited by the weakest cell. If one cell is damaged, the entire battery has to be recycled. Furthermore, a number of dedicated electronic units are required to convert the battery voltage that substantially changes throughout a cycle to match the requirements, to balance cells that drift apart, and monitor thermal hotspots as well as ageing. We developed a novel circuit technology that can dynamically reconfigure a battery to perform all the tasks that are done by separate systems, increase the performance as well as efficiency, and generate almost any output voltage and current by reconfiguring the series-parallel circuit configuration of the battery cells rapidly. By changing the circuit configuration, energy can be transferred between cells, load can be shifted or balanced, the internal resistance can be varied, and the output voltage and current rating can be adjusted rapidly.

We are looking for a candidate who wants to contribute to this novel circuit technology and/or control of the many degrees of freedom and help to innovate the way we will store energy in the future.

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial)

Application link

Design and Implementation of Electric Vehicle Drive Train with Novel Dynamically Reconfigurable Battery
Faculty Advisor: Dr. Stefan Goetz

Conventional electric vehicles use a hard-wired battery and many dedicated electronic systems to convert power for the motor as well as many other systems and monitor the battery. We are developing a technology that revolutionizes the way how electric drive trains are set up and implement it into a prototype vehicle. We would like to welcome you in our team to participate in this tectonic shift and the study the new questions such a system entails.

Dependent on the aspect you want to concentrate on, you should have experience in at least one of the following topics: circuit design and layout, microcontroller/fpga programming, control, electric motors, sensors, drive-train control.

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial)

Application link

Design of Renewable-Energy Grid-Storage System Using Novel Highly Flexible Circuit Topology
Faculty Advisor: Dr. Stefan Goetz

At present, our electric energy supply is changing rapidly. Former centralized power generation from fossil and nuclear sources is amended and partly replaced by renewable energy sources with large power share but high volatility, such as wind and solar energy. Energy storage systems are a solution to average out the fluctuations and make the supply meet the demand. While electric vehicles have stimulated impressive developments in battery technology, typical electronics concepts for such systems use rather old technology with many limitations.

We will design and implement a novel circuit technology discovered at Duke that breaks with limiting conventions to form a highly flexible reconfigurable battery system that can run each sub-battery of system at its optimum point, substantially increase the performance of the system, perform many tasks conventionally done by separate units with the same electronics now, and generate any kind of DC and AC output with high quality.

Ideally, you have experience in some of the following topics: electronics, circuit design, microcontroller programming, high-performance electronics.

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial)

Application link

Novel High-Frequency High-Power Circuitry for Noninvasive Inductive Neurostimulation
Faculty Advisor: Dr. Stefan Goetz

Noninvasive neurostimulation systems such as TMS are high-performance electronic systems that can actuate neurons in the body to trigger their response and introduce artificial signals into neural processing circuits. As the activation dynamics of neurons is type-specific and strongly nonlinear (chaotic in the mathematical sense), variation of the dynamics of the excitation pulse allows preferable activation of some neuron types and an analysis of the dynamics of the activated neurons, similar to a filter circuit analysis with Fourier sweeps in analog electronics.

Despite a world-wide desire to have a noninvasive magnetic stimulation device that can freely synthesize almost any pulse shape, there has not been an appropriate circuit technology for this challenge. We recently developed a novel circuit technology that can generate almost any output signal with high frequency content at high power as required here. We would like to welcome you in our team for participating in a development at the forefront of electronics in neuroscience and medicine.

Ideally, you have experience in at least two of the following to design an appropriate subproject according to your needs and interest: circuit design as well as layout, high-performance electronics, control systems, microcontroller and/or FPGA programming.

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial)

Application link

Signal Processing and Feature Extraction from Neurosignals
Faculty Advisor: Dr. Stefan Goetz

Neurons communicate electrically. These subtle signals can be detected to analyze and understand the electrical behavior of neural circuits. We use latest signal-processing, estimation, and machine-learning methods to detect patterns, find signals far below the noise level, and reconstruct ongoing neural computation in a living human. The project aims at developing novel methods to extract and quantify such information. Ideally, the findings are implemented in a neuroamplifier that can become a commercial product.

You ideally have experience in advanced signal processing, matlab, and first contact to machine learning and estimation.

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial)

Application link

Optical Fluorescence Imaging of the Live Brain
Faculty Advisor: Dr. Yiyang Gong

The Gong lab is looking for a researcher to investigate neural activity using fluorescence imaging. Our goal is to image the activity of thousands of neurons simultaneously in the brain of live zebrafish and live mice. This large scale imaging will allow us to better understand the brain.

The lab has expertise in creating protein sensors that respond to neural activity (see lab homepage: http://neurotoolbox.pratt.duke.edu). We are looking for a researcher to acquire the fluorescence movies from live animal samples and then process the resulting data. The research should be experienced with labview control software and/or fluorescence image processing. Projects will be geared toward the experience level of the student. Long-term effort on the project should lead to significant research products that count toward thesis work and/or publications.

Required Skills: Labview; Matlab; Fourier transforms; Signals and systems

Deadline: September 8, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study, Paid ($15/hr)

Expected Duration: One semester with potential to extend

Additional Application Documents: Most recent transcripts

Application link

Deep Learning Robotic Multifingered Grasps
Faculty Advisor: Dr. Kris Hauser

The goal of this project is to learn complex robot manipulation strategies using deep learning and massive amounts of simulation data. The student will program scripts to generate simulated grasps of objects both isolated and in clutter. Using the generated datasets, convolutional neural network architectures will be trained and evaluated to predict grasping poses and grasp pose success from RGB-D images of the object. If successful, the student will continue to evaluate the model applied to a real robot.

Required Coursework: Machine learning

Preferred Skills: Tensorflow and Python experience preferred

Deadline: September 15, 2017

Compensation Options: MS Project Exam, Independent Study (Must enroll by September 8, 2017)

Expected Duration: One semester with potential to extend

Additional Application Documents: 

Application link

Benchmarking Datasets for Volumetric and Semantic Segmentation of Indoor Scenes
Faculty Advisor: Dr. Kris Hauser

Existing 3D object segmentation and recognition algorithms only identify the structure of visible surfaces, but for many applications in robotic manipulation and augmented reality, it is important to also understand what is behind or beneath the surface. To help develop and benchmark new volumetric object segmentation algorithms, this project aims to produce a dataset of 3D indoor scenes with labelled volumetric information. The student will develop camera rigs and 3D scene capture methodologies, and then post-process and deploy the datasets on a public website.

Required Coursework: Signal processing, Computer architecture

Required Skills: C++ experience (Required), CUDA experience (Preferred)

Deadline: September 15, 2017

Compensation Options: MS Project Exam, Independent Study (Must enroll by September 8, 2017)

Expected Duration: One semester with potential to extend

Application link

Measuring the State of Health of the Body Using Microwaves
Faculty Advisor: Dr. Bill Joines

Using less than 0.02 W of power at 915 MHz (comparable to cell phone output) we can measure the composite electrical properties of the path across the body between antennas using a vector network analyzer.  Comparing these results with those obtained with human volunteers, and those obtained with static models composed of water and muscle and fat tissues, we want to be able to establish measurable guidelines for the state of health of the body in terms of the proper balance between the percentage of water, muscle and fat.  The results to date are promising, but more measurements and calculations are needed.

Required Skills: Some knowledge of electromagnetics and vector network analyzers

Deadline: October 1, 2017

Compensation Options: MS Thesis Exam, MS Project Exam

Expected Duration: One semester with potential to extend

Application link

Controlling Microwave Reflections from Conductive Surfaces
Faculty Advisor: Dr. Bill Joines

We have extended some work that was started more than 70 years ago; namely, the Jaumann shield and the Salisbury screen. These involve placing thin films, of precise conductivity, at quarter wavelength intervals in front of and parallel with a highly conductive surface to prevent or minimize reflections from that surface. We have developed the design theory that yields a maximally-flat response for any number of film layers, where the frequency bandwidth for negligible reflections increases with the number of layers. But beyond that, by adjusting the conductivity and spacing of the film layers, we can achieve better than 10 dB return loss (10\% reflected power) over the entire frequency range from 1 to 100 GHz (or 0.3 m to 3 mm wavelength). We believe that these results could be extended to even higher frequencies if needed. We also believe that materials can be found or developed that would allow us to create flexible layered materials that could be used to render military vehicles and installations (or anything else) undetectable by radar. This would include the equipment of soldiers on the battlefield or the soldiers themselves. Zero reflection would not generally be required since that state could be detectable by radar, but just enough reflection to blend into the surrounding background. Our plans for this work are to optimize the bandwidth for a desired level of return loss; for example, at 10 dB, 15 dB and 20 dB of return loss, or anywhere between 10\% and 1\% reflected power as needed for the particular application.

Required Coursework: Electromagnetics, Microwaves

Deadline: October 1, 2017

Compensation Options: MS Thesis Exam, MS Project Exam

Expected Duration: One semester with potential to extend

Application link

Lowering the Operating Frequency of Patch Antennas
Faculty Advisor: Dr. Bill Joines

A microstrip patch antenna usually consists of a rectangular or square sheet of copper on a dielectric substrate on top of a ground plane.  An input coaxial line protruding through from ground plane to patch is usually connected  at 45 electrical degrees from one end of the patch and 135 degrees from the other.  Thus, the length of the patch is 180 degrees or one half-wavelength at the chosen operating frequency.  The coaxial connection is placed at tlhe center of the patch in the width direction.  To loweer the operating frequency the patch must be made physically longer, so that the half wavelength distance is longer.  By experiment and analysis there are otther interesting ways to create the electrical delays (45 degrees and 135 degrees).that lower the frequency without increasing the patch size.  We will be exploring these ways.

Required Coursework: Electromagnetic transmission circuits

Required Skills: Some knowledge of vector network analyzers and spectrum analyzers

Deadline: October 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam

Expected Duration: One semester with potential to extend

Application link

Managing Datacenters with Game Theory
Faculty Advisor: Dr. Ben Lee

Performance-centric management policies are insufficient for privately shared computer systems. Conventional wisdom assumes that users must share and policies need only mitigate performance losses. Such performance goals are suitable for public systems that deliver hardware for which users have paid. In contrast, private systems consist of users who voluntarily combine their resources and subscribe to a common management policy. However, these users also reserve the right to withdraw from the system if resources are managed poorly.

We use algorithmic game theory to manage large, complex computer systems. Game theory is a framework for reasoning about selfish users and modeling the competition for resources. Privately shared systems must manage resources fairly to encourage participation and guard against strategic behavior. Real-word users are selfish and rational, an observation that has motivated numerous game-theoretic perspectives on systems management. Neglecting users' preferences or fairness induces strategic behavior. Users may circumvent policies or break away from shared clusters, redeploying hardware to form smaller, separate systems. Fairness addresses these challenges, ensuring system integrity and stability.

We seek research assistants with expertise in either computer systems or microeconomics / game theory, and a willingness to learn about the other discipline. Research assistants should have some experience in programming (e.g., C, C++, Java) as well as comfort with algorithms and mathematics.

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Application link

Managing Datacenters with Machine Learning
Faculty Advisor: Dr. Ben Lee

Performance anomalies threaten performance from datacenter-scale computation. Datacenters split a job into many tasks, execute them in parallel across many machines, and aggregate results when the last task completes. Stragglers are exceptionally slow tasks within a job that significantly delay its completion. Unfortunately, stragglers' effects increase with the number of tasks and scale of the system. In a Google datacenter, we find that stragglers degrade performance in 20% of jobs by more than 1.5x. We apply statistical methods in machine learning to understand system performance. These methods infer stragglers' root causes by (i) modeling performance as a function of system predictors and (ii) discovering recurring and interpretable causes of poor performance across jobs and machines.

We seek research assistants with expertise in either computer systems or machine learning, and a willingness to learn about the other discipline. Research assistants should have some experience in programming (e.g., C, C++, Java) or statistical computing (e.g., R).

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Application link

Customizing Hardware for Datacenters
Faculty Advisor: Dr. Ben Lee

Custom hardware can improve energy efficiency by 100x. General-purpose hardware is inefficient because of overheads from data movement (e.g., through the cache hierarchy) and instruction control (e.g., dynamic scheduling). Specialized accelerators eliminate these inefficiencies by tailoring the hardware to the exact needs of the software.

However, customizing hardware is expensive. Design is difficult and costs are high. Recent advances in high-level synthesis permit users to express computation in user-friendly languages, which resemble C, and automatically generate hardware descriptions with Verilog or VHDL. High-level synthesis allows the user to explore a large number of different designs for the same computation. The exploration can optimize key objectives such as performance, power, and area. We are studying new methods for (i) optimizing performance and power during design and (ii) allocating and scheduling resources in custom or reconfigurable hardware.

We seek research assistants with expertise in computer architecture and digital design. Research assistants should have some experience in Verilog or VHDL, some experience with FPGAs, as well as some comfort with programming (e.g., C, C++).

Deadline: Open until filled

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study

Expected Duration: One semester with potential to extend

Application link

AI Design of Wireless Devices
Faculty Advisor: Dr. Qing Liu

This project will apply artificial intelligence (AI) techniques for the design optimization of wireless devices, such as antennas and filters.  The student should be familiar with Matlab programming, and C++ or Java programming will be also helpful but not required.

Required Skills/Coursework: 

Deadline: January 31, 2018

Compensation Options: MS Project Exam

Expected Duration: Two semesters

Application link

Ensuring Medication Compliance and Reconciliation in Rural Population
Faculty Advisor: Dr. Miroslav Pajic

In medicine, compliance (or adherence, capacitance) describes the degree to which a patient correctly follows medical advice. Although the term applies to other situations (e.g., medical device use, self-care), most commonly it refers to medication or drug compliance. Non-compliance is a major obstacle to the effective delivery of health care. Efforts to improve compliance have been aimed at simplifying medication packaging, providing effective medication reminders, improving patient education, and limiting the number of medications prescribed simultaneously.

One of the major problems we face in rural healthcare is medication adherence, especially with patients having multiple medications at home prescribed by multiple providers. These patients do not have a method to accurately report to the physician what medications they are taking. Due to HIPAA, the primary care provider computer, other service providers’ computers and the pharmacy computer cannot talk. Patients would greatly benefit if a method can be established tracking all of their prescriptions, reminding them to take the right medicine at the right time, understand which medicines are for what conditions, and which medicines should never be taken at the same time - despite the fact that they have been prescribed to them (by non-communicating providers). In addition, they need a way to know if they have already taken their medication. This is especially challenging in rural areas, where cell service and internet are not always available. These patients would benefit from a mobile repository of all their prescribed medications available to all healthcare providers involved in their care. The benefit would be even greater if drug interactions and warnings could be generated by the system, as well as reminders and possible confirmation of taking prescribed doses in prescribed intervals.

The goal of this project is to evaluate HIPAA and safety requirements for the aforementioned system and to develop and test an initial prototype in collaboration with West Virginia School of Osteopathic Medicine, Lewisburg, WV. Specific challenges that needs to be addressed include: a) How to provide privacy and security guarantees? b) How can we ensure fault tolerant operation in the presence of limited communication infrastructure, with intermitted network access (using various communication standards)? c) How can we ensure that the developed technology takes into account limited technical understanding of some of the users?

Required Skills/Coursework: Any relevant or prior experience 

Deadline: October 1, 2017

Compensation Options: MS Thesis Exam

Expected Duration: Two semesters

Application link

Tracking and Coordinating Pediatric Care for Neonatal Abstinence Syndrome Infants
Faculty Advisor: Dr. Miroslav Pajic

Due to an ongoing opioid epidemic, the number of babies suffering from neonatal abstinence syndrome is increasing. Fetal and neonatal addiction and withdrawal are a result of the mother's dependence on drugs during pregnancy. Neonatal abstinence syndrome patients require continued multidisciplinary care. At a minimum, professionals involved in this care include obstetricians, neonatologists, pediatricians, nurses, nutritionists, mental health professionals, social workers, substance abuse counselors, and child development specialists.

Due to geographic and economic disadvantages, multidisciplinary treatment centers are not readily available to rural patients. Keeping track of all the required services, compliance and follow-ups would be in the best interest of these children. Furthermore, keeping track of all the services and appointments required for these patients is a challenge under normal circumstances and with a supportive family. However, more often than not, these infants are being transferred between multiple state system foster care providers and the biological parents for prolonged periods of time. The primary care providers, as well as other services, tend to change with each geographical and/or legal guardian change. This adds an unnecessary level of burden to providing care for these children and may contribute to developmental delays. It would be extremely beneficial to have a personalized central repository of all the relevant health records, ongoing and required medical services, as well as progress notes that would follow the child regardless of their geographical location or foster (parent) and medical provider status. Without such a repository, there is currently no way to keep the continuum of care for this child by a newly established primary care provider or a social worker.

The goal of this project is to evaluate HIPAA and safety requirements for the aforementioned system and to develop and test an initial prototype in collaboration with West Virginia School of Osteopathic Medicine, Lewisburg, WV. Specific challenges that needs to be addressed include: a) How to provide privacy and security guarantees? b) How can we ensure fault tolerant operation in the presence of limited communication infrastructure, with intermitted network access (using various communication standards)? c) How can we ensure that the developed technology takes into account limited technical understanding of some of the users?

Required Skills/Coursework: Any relevant or prior experience 

Deadline: October 1, 2017

Compensation Options: MS Thesis Exam

Expected Duration: Two semesters

Application link

Securing Autonomous Systems in Contested Environments
Faculty Advisor: Dr. Miroslav Pajic

Most of existing control and autonomous systems have not been built with security in mind. Even with the proliferation of different networking technologies and the use of more open control architectures, until recently their security has usually been an afterthought. Yet, with the advance of cyber-physical systems (CPS), the tight interaction between information technology and the physical world have made control components of CPS vulnerable to attack vectors well beyond the standard cyber attacks. In the last few years, several incidents have clearly illustrated susceptibility of CPS to attacks, raising attention to serious security challenges. These include the StuxNet virus attack on an industrial SCADA system, as well as attacks on modern cars and RQ-170 Sentinel US drone that was captured in Iran.

Relying exclusively on cyber-security techniques for securing CPS is insufficient. For example, GPS spoofing attacks to misguide a yacht or unmanned aerial vehicle (UAV) of their routes have been demonstrated in recent years. Consequently, the goal of this project is to show how combination of security-aware control policies can be used to increase security guarantees in autonomous systems operating in contested environments. The scope of this project will be to secure the video-based navigation of an UAV, by exploiting recently developed control techniques such as attack-resilient state estimation and active attack detection.

Required Skills: C, C++, knowledge of ROS is a plus 

Deadline: October 1, 2017

Compensation Options: MS Thesis Exam

Expected Duration: Two semesters

Application link

Ensuring Timeliness in Safety-Critical IoT Systems
Faculty Advisor: Dr. Miroslav Pajic

To manage complexity of modern Internet-of-Things (IoT) systems, employed platforms require the use of efficient operating systems (OS) that provide timing, performance and security guarantees, as well as support for key IoT networking and communication technologies. One example is the latest mbed OS 5.1, which incorporates a Real-Time Operating System (RTOS) in order to provide native real-time thread support to the applications running on top of the RTOS.

The RTOS core is based on the widely used open-source CMSIS-RTOS RTX - an established kernel that can support threads and other RTOS services on very resource constrained devices. However, the underlying scheduling is based on thve well known round-robin policy, effectively preventing the use of this OS in development on safety critical IoT applications where timeliness is the first class citizen. In these systems (e.g., in industrial, medical, automotive, avionics and many other application domains), the ability to prioritize a critical task at any given moment in time, and guarantee availability of resources required to complete that task is essential.

Consequently, the goal of this project is to extend the open-source scheduler for the RTOS and provide support for a variety of commonly used real-time scheduling policies that take into account thread criticality and execution deadline constraints. The developed RTOS will then be deployed and evaluated in several case studies in automotive, industrial automation, and medical device domains, allowing the student to both learn foundations of the RTOS development as well as its use in design of real-world systems (e.g., electric vehicles, cloud-based reconfigurable industrial systems, networked implantable medical devices).

Required Skills: C, C++, knowledge of embedded systems 

Deadline: October 1, 2017

Compensation Options: MS Project Exam

Expected Duration: One semester with potential to extend

Application link

Software Design for Autonomous Formula F1/10 vehicles
Faculty Advisor: Dr. Miroslav Pajic

The F1/10 competition focuses on creating a meaningful and challenging design experience for students. The competition involves designing, building, and testing an autonomous 1/10th scale F1 race car (capable of speeds in excess of 40MPH) all while learning about perception, planning, and control for autonomous navigation. The students in Duke team would start from the initial vehicle design and ROS-based software that provides baseline autonomous vehicle operation. More information available at: http://f1tenth.org

Required Skills: C, C++, Knowledge of ROS is a plus 

Deadline: October 1, 2017

Compensation Options: MS Project Exam

Expected Duration: One semester with potential to extend

Application link

UPP2SF Tool development
Faculty Advisor: Dr. Miroslav Pajic

Model-Driven Design (MDD) of cyber-physical systems advocates for design procedures that start with formal modeling of the real-time system, followed by the model’s verification at an early stage. The verified model must then be translated to a more detailed model for simulation-based testing and finally translated into executable code in a physical implementation. As later stages build on the same core model, it is essential that models used earlier in the pipeline are valid approximations of the more detailed models developed downstream. The focus of this effort is on the design and development of a model translation tool, UPP2SF, and how it integrates system modeling, verification, model-based WCET analysis, simulation and code generation into an MDD-based framework. UPP2SF should facilitate automatic conversion of verified timed automata-based models (in UPPAAL) to models that may be simulated and tested (in Simulink/Stateflow). The existing tool revision should be based on the design rules to ensure the conversion is correct, efficient and applicable to a large class of models.

Required Skills: Matlab

Deadline: October 1, 2017

Compensation Options: MS Project Exam

Expected Duration: One semester with potential to extend

Application link

Coils for Transcranial Magnetic Stimulation (TMS)
Faculty Advisor: Dr. Angel Peterchev

TMS is a tool for non-invasive modulation of brain function for research and therapeutic purposes (FDA approved for depression since 2008). This project involves design, implementation, and testing of advanced electromagnetic coils for TMS. Tasks involve using CAD software to design the coils, mechanical construction, and electrical and acoustic testing and characterization. The project provides an opportunity to interact with an interdisciplinary team of researchers.

Required Skills: Hands on skills with CAD, 3D printing, machining, electrical measurements

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study (must be enrolled by Septeber 8, 2017), Paid ($15-$20/hr, dependent upon skill level)

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial) and three professional references.

Application link

Rapid-Recovery Bioamplifier
Faculty Advisor: Dr. Angel Peterchev

In brain stimulation research and clinical applications there is increasing need to record brain signals (electroencephalography, local field potentials, single neuron spiking) immediately after electric or magnetic stimulation of the brain. However, the electromagnetic artifact corrupts the neural recording after the stimulus pulses. The objective of this project is to develop a neural amplifier that recovers rapidly (< 1 ms) after a stimulus pulse, opening window into the neural dynamics in the immediate aftermath of stimulation. Tasks involve specification, circuit design, simulation, implementation, debugging, and testing of a rapid recovery bioamplifier. The project provides an opportunity to interact with an interdisciplinary team of researchers.

Required Coursework: Microelectronic circuits or equivalent

Required Skills: Hands on experience with electronic circuit design, simulation, construction, and testing

Deadline: December 31, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study (must be enrolled by Septeber 8, 2017), Paid ($15-$20/hr, dependent upon skill level)

Expected Duration: One semester with potential to extend

Additional Application Documents: Undergraduate transcript (can be unofficial) and three professional references.

Application link

Community Detection with Structured Stochastic Block Models
Faculty Advisor: Dr. Galen Reeves

The problem of community detection is to find important clusters in a network. Specific examples include finding like-minded people in a social network and discovering the hierarchical relationships in organizations from observed behavior. The state-of-the art algorithms for this problem are based on spectral clustering techniques. The goal of this project is to study improved algorithms based on a divide-and-conquer approach. The work will require the implementation of the new approach and comparison with existing methods on both synthetic and representative real-world datasets.

The project is appropriate for a student who has a strong background in linear algebra (ECE 590), probability (ECE 581), and machine learning (ECE 681 or STA 561). The work will require programming in MATLAB. The scope of the project is one to two semesters.

Deadline: September 22, 2017

Compensation Options: MS Thesis Exam, MS Project Exam, Independent Study (must be enrolled by September 8, 2017

Expected Duration: One semester with potential to extend

Application link