GAANN Fellowship in Electrical Engineering and Computer Science
Top students in STEM fields pursuing doctoral degrees may be eligible for Graduate Assistance in Areas of National Need (GAANN) Fellowships in Electrical Engineering and related applications. These Fellowships are awarded based on academic performance and financial need. GAANN Fellows are recommended for the program by the School of Electrical Engineering and Computer Science and require approval by the Graduate School and the Fellowship Program Selections Committee. If selected, GAANN Fellows are eligible for stipends up to $34,000 per year, including a tuition waiver.
Enrolled full-time in or admitted to the doctoral program in Electrical Engineering and Computer Science
U.S. citizen, permanent resident, or a permanent resident of a Free State
Committed to a career as a university faculty member or high-impact researcher
Outstanding undergraduate and (if applicable) graduate academic record (cumulative grade point average)
Demonstrated financial need, determined according to federal guidelines
How to Apply
Step 1. Apply to the Ph.D. program in the School of Electrical Engineering and Computer Science. Qualified students may apply for the Ph.D. program with B.S. degree (direct entry).
Step 2. Email Dr. Wojciech Jadwisienczak at firstname.lastname@example.org expressing your intent to apply for the GAANN Fellowship.
GAANN Faculty and Department Affiliations
- Wojciech Jadwisienczak, Professor, Electrical Engineering
- David Juedes (Director), Professor, Computer Science
- Avinash Karanth, Professor, Electrical Engineering
- Savas Kaya, Professor, Electrical Engineering
- Chang Liu, Professor, Computer Science
- Jundong Liu, Associate Professor, Computer Science
- Chad Mourning, Assistant Professor, Electrical Engineering
- Faiz Rahman, Associate Professor, Electrical Engineering
The GAANN Fellowship program is open to all qualifying students interested in intensive research at the graduate level and teaching activities. Ph.D. in Electrical Engineering and Computer Science at Ohio University prepares students to take on a wide variety of research tasks surrounding electromagnetism, electricity, electronics, and novel information technologies. Courses deliver breadth and depth of knowledge on principles and fundamental concepts of Electrical Engineering, Physics, Mathematics, Computer Science, Electromagnetism, and novel Electronics. The GAANN awardees will pursue required coursework, a three-part comprehensive examination, and independent research, and will be trained to become successful academic instructors, mentors, and researchers. Our coursework reflects the diverse expertise of our faculty and addresses the current trends and technologies shaping the profession today. The GAANN students will have the opportunity to enhance their class experience when joining faculty research focused on solving real-world problems. Research in the School of EECS includes world-class research in EE areas of avionics and navigation, communications and signal processing, controls, computer engineering, computer architecture, networking, electromagnetics, opto- and nano-electronics, industrial controls, and analog and digital circuits, and in CS areas of algorithm design, theory of computation, software verification and certification, security, programming languages, computer networking, artificial intelligence, robotics, bioinformatics, visualization, and image processing. Please visit the EECS Faculty websites for more information on research projects available or select from the list of research areas shown below.
Boron Nitride Explored for Intelligent Sensors Platform
(Jadwisienczak and Rahman)
Atomically thin hexagonal boron nitride (BN) single layer and nanosheets (BNNSs) have been widely explored for various applications due to their unique properties including electrochemical sensing gas molecules with high sensitivity and selectivity, sensing of the ultra-low magnetic field, fluorescence sensing of DNA and extensive DNA related analytes and refractive index sensing to identify biomolecules and chemicals in the mid-infrared range for drug discovery, bioengineering, and environmental monitoring. The excellent electrocatalytic activity, high specific surface area, N- and B-active edges, structural defects, adjustable band gap through interaction with other nanomaterials, and chemical functionalization, make 2D h-BN ideal for developing an intelligent sensing platform. Furthermore, the BNNT can be used to identify both qualitative and quantitative detection of contaminants. In this project directed by two faculty with expertise in material science (Dr. Jadwisienczak) and device engineering (Dr. Rahman), a prospective GAANN fellow with explore finding a pathway to modulate the electronic properties of 2D BN while the intrinsic characteristics are well preserved, to design, fabricate and characterize a new generation of highly sensitive BN-based sensors. The student will explore the fabrication techniques of various 2D BN allotropes, followed by a thorough exploration of their advantages, shortcomings, and promising possibilities as sensing platforms in the project. It is anticipated that a student fellow will conduct extensive research focused on selected tasks relevant to (1) the challenge to functionalize 2D BN by controlling the surface chemical reactions with external species, particularly metal nanoparticles, (2) the encapsulation of h-BN with graphene nanoribbon or other selected 2D materials to make the electronic system ultraclean, and electrically tunable for achieving refractive index sensor with ultrahigh sensitivity, (3) engineer a plasmon resonance featuring a higher-quality factor in the graphene/h-BN few-layer ribbon array and (4) ultrathin graphene and h-BN sandwich structure as magnetic field Hall-effect sensor detecting minuscule changes in magnetic fields at different temperatures. Furthermore, GANN fellow may participate in a study of how deep learning algorithms can enable predicting the physical properties of the selected BN-based material system configuration.
Machine Learning Applications for Determining Visibility in Aviation (Mourning)
Visibility in aviation is defined as the greatest distance through the atmosphere, toward the horizon, that prominent objects can be identified with the naked eye. Visibility is a driving component behind many aviation incidents and accidents and, therefore, aviation regulations. In order to provide the necessary data to comply with these regulations, visibility sensors exist in the form of atmospheric transmission-meters and optical scattero-meters, but the price point on these sensors makes them infeasible for high fidelity (one data point per square mile) measurement networks on the order of a state or nation. The objective of this proposal is to develop and evaluate a low-cost visibility sensor using the combination of more commonly (and cheaply) available electro-optical sensors and machine learning algorithms. By using existing sensors, truth data can be created that pair visibility values, measured in the US in statue miles, with images of the atmospheric conditions that yielded that result, possibility coupled with other more easily (and cheaply) measured weather criteria such as temperature, humidity, and time of day. This dataset fits well-established paradigms in training regression networks and merits further exploration. Additional research may include expansion to stereo-camera or wide-spectrum approaches incorporating IR and UV into the approach, as well as predicting changes in visibility state in the future based on time series data.
Development of Novel Flexible Multimodal Sensors for Sensor Fusion (Kaya, Rahman, Jadwisienczak)
The last decade of research on flexible and printed electronics has paved the way for sensor-enriched, smart, wireless devices that can be miniaturized into stamp-size elements and embedded into everyday objects. Sensors truly ‘flexible’ in terms of both functionality and form are required to fully explore numerous exciting and novel applications. In particular, in situations where the collected data is to be used for critical decisions such as health and safety, correlated inputs from multiple sensors fused together in realistic operating conditions are needed. Thanks to a wide range of developments in nanomaterials, low-cost printed electronics, and sensor design, our team has been an active participant of this phenomenal growth in multi-modal sensor development and fabrication, especially in terms of novel printed passives and electro-optic sensors on both conventional semiconductor and flexible (paper, polymer, and textile) substrates. Our current activities encompass a wide range of sensor materials, design and fabrication techniques that can be applied to both biomedical and environmental sensing in terms of electrical, magnetic, optical, chemical, and mechanical inputs. Hence, we are in a unique position to develop and introduce many flexible sensors that will enable the current initiative to make an even larger impact while also formulating solutions for building novel sensors and analyzing their responses. For the GAANN program, we plan to focus not only on broadening the types of sensors via sol-gel-based metal oxide (ZnO, IGO, IGZO) thin-film and printed organic (small-molecule Pentacene and polymer PANI) transistors with gain but also on creating arrays of printed sensors that can ‘map-out’ signals on a surface or distributed network. Given the multiplicity in the number and type of sensing vectors available to us, we can develop platforms especially important for sensor fusion, which requires the presence of multiple sensors with complementary and overlapping capabilities. In sensor fusion, the focus shifts from the development of very-high accuracy sensors with expensive specialized materials and complex circuitry to a variety of sensors as well as better signal processing, statistical modeling, and machine learning for more reliable decisions from multi-modal and low-cost sensors in a given context. Since flexible/printed electronic sensors are known for their very low cost rather than ultra-high performance, they blend really well with this paradigm shift. Focus on sensor fusion will also allow students to interact with the other EECS Faculty to develop algorithmic and embedded solutions. A particular emphasis will be placed on biomedical sensors for wearable monitors for elderly and chronic patients and navigational/proximity sensors that are already of interest to many faculty members in the school of EECS.
Enforcing Runtime Secure Policies by Sensing Threats in Embedded Systems (Karanth, Juedes)
Embedded systems in the form of microcontrollers and programmable logic controllers (PLCs) form most of the computing infrastructure that controls the industrial processes and critical infrastructure that our economy depends upon. These low-powered embedded systems are vulnerable to external attacks when malicious code (e.g. in the form of control-flow injection) implants itself into program memory and bypasses conventional hardware protection schemes. The feasibility of such attacks became evident when it was revealed that the STUXNET virus purportedly damaged over 1000 Iranian centrifuges] by causing those centrifuges to speed up and slow down beyond their normal operating ranges. Similar attacks could cause widespread havoc in critical infrastructure in the U.S. The focus of the proposed research is to prevent code-injection attacks by designing embedded that enforce specified logical constraints in hardware. We posit that the fundamental flaw in current embedded systems is that arbitrary code can be executed. Therefore, secure policies (rules) where programs are paired with certificates of correctness (e.g., mathematical proofs) that guarantee that given operational constraints are satisfied will prevent cyber-attacks such as Stuxnet. The hardware would not run the corrupted program because it would detect that the program did not provide the associated evidence that they met the appropriate operational constraints. Since enforcing policies (rules) in software is prohibitively expensive, prior work such as in PUMP implements partial policies statically in hardware. However, such static analyzers are incomplete since all program properties cannot be decided at compile-time and could potentially incur higher costs in implementing the security policy. In this research, we propose to design a high-level dynamic language/policy that facilitates the verification and implementation of a large class of hardware runtime monitors which can adapt to program behavior by runtime reconfiguration. To achieve our goal, we leverage expertise in three areas (language design, software/hardware verification, and computer architecture/security) to build a toolchain for specifying and implementing correct-by-construction hardware security runtime monitors. By leveraging prior work where PIs designed GARUDA, we propose to develop a new language, Dynamic-GARUDA that will facilitate the verification and implementation of a more expansive class of hardware runtime monitors that can sense hardware threats.
Low SWaP-C autonomous drones guided by intelligent sensors (J. Liu)
Unmanned Aerial Vehicles (UAVs) or drones are currently being actively developed and utilized for new tasks, such as surveillance, transportation, and security, among others. Fully automated drones will bring efficiencies to fuel the economy in agriculture, mining, and many manufacturing segments. Training autonomous drones: This project aims to apply deep reinforcement learning to fly autonomous drones, guided by intelligent sensors, in a safe and efficient manner. Both global and local path planning of the drones will be first trained in simulation environments to learn how to autonomously navigate to a target location without human commands or other automated controls. Desired flying styles, including smooth, fast, and collision-free, will be effectively enforced in training the flight policy. When deployed onto real drones, signals from visual (cameras) and laser sensors (LiDARs) will be seamlessly fused to provide depth maps, scene segmentations, and location information for the learned policy to make flight decisions. Low SWaP-C implementations: Neural networks implemented on general-purpose processors, such as central processing units (CPUs) and graphics processing units (GPUs) commonly have high energy and computational requirements. In this project, we plan to develop our autonomous drone solutions into low size, weight, power, and cost (SWaP-C) implementations on neuromorphic hardware. This hardware design paradigm, inspired by biological brains, has demonstrated great effectiveness in achieving low SWaP-C in real-time systems.
Verbatim Model Manifestation in Manageable Neighborhoods for Explainable Machine (Ch. Liu)
Machine learning models are typically designed and fine-tuned for optimal accuracy, which often results in layers of weights that are difficult to explain or understand. In the meantime, recent successes of machine learning systems have attracted adoption from more end users, who need to better understand the model to trust or properly use such machine learning systems. To make these two ends meet, researchers and practitioners alike have adopted several approaches, including (1) using approximate models just for an explanation; (2) linear local explanation for complex global models (e.g. LIME); (3) example-based explanation by finding and showing most influential training data points. These approaches all have their own merits, but none of them deliver everything needed by end users. The fundamental limitation of these approaches is that they assume that (1) certain aspects of machine learning systems, especially complex deep neural networks, cannot be understood by human beings; and (2) typical human users can only understand simple concepts such as linear systems or decision trees. We aim to improve on previous attempts with two assumptions. First, human users are intelligent, just not in the same way as machines. Humans can identify patterns intelligently but may not be able to scale up to thousands of data points easily. Second, machine learning systems are built to reflect actual physical systems that follow logical physical rules. What worked well most likely can be explained, even though the explanation could be complex. What cannot be explained most likely is not a good reflection of the underlying physical properties. In this work, we proposed a simple yet effective framework to help human users better understand the prediction of a black-box machine learning model. Our framework aims to build a bridge between machine learning models and human intelligence to address the machine learning model explainability problem, particularly for end users. We intend to make improvements in this area by (1) presenting various aspects of the actual model through verbatim model manifestation (instead of trying to approximate the models), and (2) identifying and generating a manageable number of data points to present to users in the local context of the data point of interest, so that software engineers and users alike can use their own intelligence to understand what the actual model is trying to do within a limited scope that is manageable by a human being.