Doctor of Philosophy, The Ohio State University, 2019, Industrial and Systems Engineering

Edge fracture, defined simply as fracture originated from the edge, is a common problem in sheet metal forming, especially when forming Advanced High Strength Steels (AHSS). The increase of AHSS in the automotive industry has derived into substantial efforts in order to predict and prevent the edge fracture phenomenon. Nearly every sheet metal forming operation includes cutting of the material at some point and, depending on the cutting process, the damage produced at the edge reduces its formability in subsequent operations. The edge fracture has been approached from various perspectives: observation of the cut edges in order to correlate their geometries to edge fracture occurrence, methods to evaluate edge stretchability, Finite Element (FE) simulations to predict it, design of new cutting methods (i.e. tool geometries or configurations) or post-processing of the edge to eliminate residual stresses (i.e. annealing or machining). Nevertheless, there is not a consensus in the field regarding which ones are the most important factors that should be considered in sheet metal cutting. Furthermore, researchers, very often, overlook practical parameters during laboratory test (e.g. non-uniform cutting clearances, tool wear, etc.); hence, these laboratory tests, generally, do not represent the cutting conditions found in the practice. This project aims to propose guidelines that could be considered to evaluate, maintain and improve a given cutting process for a required edge stretchability. In order to achieve this objective, four cases studies, which include most of the of the typical challenges in sheet metal cutting, were conducted: trimming along a straight line, piercing of a round hole, blanking of an irregular geometry and use of shaving (two-stage cutting) to improve edge stretchability. Chapter 1 of this study gives a brief introduction to the edge fracture problem and the available sheet metal cutting techniques. Chapter 2 summarizes the research objec (open full item for complete abstract)

Master of Science, University of Toledo, 2023, Electrical Engineering

Modern industrial automation utilizes local controllers based on microcontrollers, programmable logic controllers, remote terminal units, and edge servers to form an Edge-assisted Networked Control System (EaNCS). The edge server offers high computational capacity at the cost of network delays and potential data loss over wired/wireless networks, whereas local controllers provide reliable performance. This thesis focuses on the delay analysis of the EaNCS. The network delay is analyzed with the help of Network Calculus. Computational delays in local controllers and edge servers are modeled by measurements. The network and computational delays are parameters to design guidelines that we propose, which answer the question of how many local controllers an edge server can support considering the traffic profile and network capacity in an EaNCS. The design guidelines have been experimentally verified in a physical testbed that consists of an edge server, a network switch, Raspberry Pis, and a programmable logic controller.

Master of Science (MS), Wright State University, 2021, Computer Science

Today there is a large market for Unmanned Aerial Systems. Although most current systems are remotely piloted by operators on the ground, increasingly, many of these systems will use some sort of automatic flight controller to help mitigate new challenges, due to their deployment at growing scale. These challenges include, but are not limited to, shortage of FAA-certified UAS pilots, transmission bandwidth and delay constraints and cyber security threats associated with wireless networking, profitability of operations constrained by energy capacity and efficiency and air dynamics planning, and etc. In order to address these rising challenges, this thesis is a part of an effort to develop and test an on-board Sense and Avoid system for assisted and/or autonomous UAS operations. In particular, this work focuses on applying OpenCV and established computer vision algorithms to implement an object detection capability, which is a critical component of an Adaptive Two-Stage Edge-Centric Sense-and-Avoidance system. Additional efforts were made for integrating this capability into the overall system operations. Furthermore, two implements of the detection system are completed: one in C/C++ and the other in Python with an aim to compare their efficiency. It is found that both implements meet the real-time operation requirements, and experimental studies show little to no difference in processing time for object detection.

Doctor of Philosophy (PhD), Wright State University, 2021, Computer Science and Engineering PhD

With the growing number of Unmanned Aircraft Systems, current network-centric architectures present limitations in meeting real-time and time-critical requirements. Current methods utilizing centralized off-platform processing have inherent energy inefficiencies, scalability challenges, performance concerns, and cyber vulnerabilities. In this dissertation, an adaptive, two-stage, energy-efficient, edge-centric architecture is proposed to address these limitations. A novel, edge-centric Sense-and-Avoidance architecture framework is presented, and a corresponding prototype is developed using commercial hardware to validate the proposed architecture. Instead of a network-centric approach, processing is distributed at the logical edge of the sensors, and organized as Detection and Classification Subsystems. Classical machine vision algorithms are used to detect and produce a region of interest. The region of interest is then segmented and fed to the Classification Subsystem to be classified using optimized neural networks. A compressed frame from the Detection Subsystem, along with the region of interest and classification results from the Classification Subsystem, can be sent to the Ground Control Station to produce an Artificial Intelligence enhanced view to increase operator comprehension. Experimentation and testing indicate this approach is feasible for real-time operations supporting throughput of at least three 4K frames per second. Additionally, on-platform detection and classification can occur without offloading large amounts of imagery to ground processing, thereby reducing unnecessary network transmissions and associated energy consumption. The sufficient processing frame rate effectively eliminates any hover during sensing processing, demonstrating how the architecture can reduce energy consumption for battery-powered electric unmanned aerial vehicles. This novel approach opens new opportunities to reduce power consumption in future electric transpo (open full item for complete abstract)

Master of Science in Biomedical Sciences (MSBS), University of Toledo, 2018, Biomedical Sciences (Medical Physics: Radiation Oncology)

The University of Toledo currently uses the 2.5mm MLC leaves on our Varian Edge (Varian Medical Systems Inc., Palo Alto, CA) for every SRS case. For most cases the MLCs provide excellent results. However, their unanimous use in the clinic leads to question if they are superior to cones for every case. For small single fraction cases, it may be possible that better plans can be achieved with the use of SRS cones. It is this topic which we investigate in a retrospective study which is a continuation of previous work done here at the University of Toledo. Varian stereotactic cones are commissioned in Raystation6 (RaySearch Laboratories, Stockholm, Sweden). Patients with small (<2.8cc) single fraction SRS targets are re-planned with the Varian cones and compared to the original MLC defined plans using the RTOG conformity index, Paddick conformity index, Conformity Gradient index, Gradient index, Dose Gradient index, and an in-house gradient index. Target volumes are also evaluated for the minimum spherical volume required to circumscribe them. This spherical index is compared with the gradient and conformity indices to investigate a potential correlation. New treatment plans are created using Raystation6 and Varian SRS cones. The plans show steeper dose fall-off can be achieved with stereotactic cones compared to MLCs. The plans showed no significant difference in dose conformity to the target volume between the modalities or any correlation of conformity to the shape of the target.

Master of Science, The Ohio State University, 2015, Environment and Natural Resources

Clean farming practices and forest succession have contributed to population decline of northern bobwhites (Colinus virginianus) across northern portions of their range. Intensively farmed landscapes lack early successional vegetation that provides protective cover near food sources. Earlier research indicated that population growth of northern bobwhites in southwestern Ohio is limited by lack of preferred early successional woody cover during the non-breeding season. I studied vegetation response to removal of large trees from wooded edges (here after edge-feathering) on private owned farmlands in Highland County Ohio. Ninety-nine areas ranging in length from 15 m to 91 m were treated during spring in 2012 and 2013. Vegetation structure and composition of feathered edges was measured before treatment and after 2 growing seasons and in late winter during 2012 – 2014. I used repeated measures analysis of variance to test for differences in vegetation structure and composition among study sites, edge aspects, feathered edge size classes, edge types, and basal area reduction. Basal area reduction differed between years, with a light reduction (29%) in 2012 and a heavy reduction (81%) in 2013. Horizontal, vertical, and ground cover differed among sample periods with the second fall having more vertical and horizontal cover than the first fall, and the first fall having more cover than the first winter. Basal area reduction, size, and sample periods were important predictors of cover measurements. Basal area reduction within woodlot edges or along linear features like fencerows was the most important variable that affected vegetation response to edge-feathering. Basal area reductions between 37 – 50% resulted in positive changes in protective cover for bobwhites after 1 growing season. Large edge treatments with heavy reduction in basal area resulted in net gains in protective cover between seasons, and provided the highest overall change in cover. (open full item for complete abstract)

Master of Science in Mathematics, Youngstown State University, 2014, Department of Mathematics and Statistics

A friendly partition of a graph is a partition of the vertices into two sets so that every vertex has at least as many neighbors (adjacent vertices) in its own set as in the other set. An unfriendly partition of a graph is a partition of the vertices into two sets so that every vertex has at least as many neighbors in the other set as in its own set. In this paper we extend these concepts to k-partitions of vertices. We define and explore friendly and unfriendly edge partitions and extend these concepts to k-partitions of edges. In extending these concepts to the edges of a graph, we will show that one type of a friendly vertex partition of a Km,n graph can be used to produce a friendly edge partition. We will also look at partitions that are both friendly and unfriendly (dual). We will investigate these properties for several types of graphs (star, tree, Kn, Cn, Km,n).

Doctor of Philosophy, The Ohio State University, 2009, Mathematics

In this thesis, we consider the problem posed by Gupta, Goldberg and Seymour: “When is the chromatic index χ′ of a graph almost the same as its fractional chromatic index χ′*?”. While they conjectured that the answer is when “the chromatic index is at least 2 more than the maximum degree of the graph”, so far this problem is wide open. The first part of this thesis improves the known results to this problem by following the work of Tashkinov and using the VKT-trees introduced by Chen, Yu and Zang. In the second part, we show that the ideas of fan and path can be generalized to the analogous problem posed by Gupta by replacing the idea of “covering with matchings” with “packing with edge covers”. While we will not present any explicit results such as the one presented in the first part, this part provides some insight into the nature of alternating chains in this new context. The last part of this thesis will be a tribute to a recent result by Goldberg. Goldberg shows that it is not necessary to use the matching number of a graph while working on Goldberg-Gupta Conjecture. We prove the analogous result for the analogous conjecture.

Doctor of Philosophy, The Ohio State University, 2006, Electrical Engineering

The distributed effects on the one-finger FET (lateral distributed effects) are analyzed and analytical solutions are obtained. A 4-port network model is reported for the 6-terminal device in the limit of small device width. Negative gate, drain and source feedback impedances are observed in the 4-port network. Two boundary conditions are considered resulting in two distributed 2-port equivalent circuits. For transistors with large periphery, distributed effects along gate and drain rails are analyzed. An analytical solution is presented in the small length approximation, including both distributed effects along the rails and device width. The distributed effects due to the rails become dominant as the finger number is increased. The parasitic network of an AlGaN/GaN FET is studied using Momentum. A Momentum simulation block for AlGaN/GaN HFETs is defined. To extract the lateral impedances, Momentum simulations are fitted using the 4-port network model. The parasitics caused by the air-bridges are obtained and the effect on the device performance is examined. The scalability of the device is examined. The Agere Electro-thermal Transistor (AET) model is modified to fit the AlGaN/GaN HFET. The DC-IV model is extracted from measured pulsed-IV for a 2-finger 150μm AlGaN/GaN. The small signal model is extracted to obtain operating frequencies of 11.7GHz and 48.2GHz. A center-fed transistor and an edge-fed transistor are investigated to implement the 11.7GHz and 48.2GHz AlGaN/GaN HFET devices. More signal distributed effects due to the rails appears in the edge-fed devices over the center-fed devices and in the high frequency devices over the low frequency devices. For a 10dB gain, the optimal devices with the maximum peripheries are obtained for each devices and their P1dB's and PAE's are compared. Several layouts -corporate tree, tapered edge, lateral corporate tree (LC), LC with bypasses (LCB)- are designed to improve the center-fed and edge-fed layouts. The methodo (open full item for complete abstract)

MS, Kent State University, 2024, College of Arts and Sciences / Department of Geography

Wildfire and drought are key drivers of shrubland expansion in southwestern US landscapes. Stand-replacing fires in dry conifer forests induce shrub-dominated stages, and changing climatic patterns may cause a long-term shift from coniferous forests to deciduous shrublands. This study assessed recent changes in deciduous fractional shrub cover (DFSC) in the eastern Jemez Mountains from 2019-2023 using topographic and Sentinel-2 satellite data in a random forest model. Sentinel-2 provides multispectral bands at 10 and 20 meters, including three 20 meter red edge bands, which are highly sensitive to variation in vegetation. There is no consensus in the literature on whether upscaling imagery to 20 meters or downscaling to 10 meters is more advantageous. Therefore, an additional goal of this study was to evaluate the impact of spatial scale on DFSC model performance. Two random forest models were built, a 10 and 20 meter model. The 20 meter model outperformed the 10 meter model, achieving an R-squared value of 0.82 and an RMSE of 7.85, compared to the 10 meter model (0.76 and 9.99, respectively). The 20 meter model, built from 2020 satellite imagery, was projected to the other years of the study, by replacing the spectral variables with satellite imagery from the respective year, resulting in yearly predictions of DFSC from 2019-2023. DFSC decreased from 2019-2022, coinciding with severe drought and a 2022 fire, followed by a significant increase in 2023, particularly within the 2022 fire footprint. Overall trends showed a general increase in DFSC despite high interannual variability, with elevation being a key topographic variable influencing these trends. This study revealed yearly vegetation dynamics in a semi-arid system and provided a close look at post-fire regeneration patterns in deciduous resprouting shrubs. Understanding this complex system is crucial for informing management strategies as the landscape continues to shift from conifer forest to shrubland du (open full item for complete abstract)

Bachelor of Arts (BA), Ohio University, 2024, Environmental Studies

Grassland birds have narrow habitat requirements that are influenced by food availability, habitat composition, and habitat structure. Because survival is influenced by habitat quality and availability, understanding habitat requirements is critical for conservation. I determined how Henslow's Sparrows (Centronyx henslowii) use grassland habitat in the breeding season. In the past 10 years, technology advancements have allowed researchers to study the habitat use and movement ecology of understudied birds, such as grassland birds. I deployed nanotags on 47 adult Henslow's Sparrows at two sites in southern Ohio to determine home range size and habitat use in relation to distance to edge and shrub, as well as the post-breeding dispersal and migratory timing. I predicted that Henslow's Sparrows would use core grassland habitat and avoid edge and shrubs. I found no difference in 95 % home range size between female (0.10 ± 0.03 ha) and male (0.32 ± 0.18 ha) Henslow's Sparrows. Henslow's Sparrows used shrubs when available and edge habitat as refugia after disturbance. I also found that Henslow's Sparrows use fields into August, past dates typically recommended for disturbance (e.g., mid to late July), which suggests the need to leave corridors and patches for refugia after management such as mowing or burning. I determined the fall migratory departure timing of 13 Henslow's Sparrows. I found that Henslow's Sparrows are at risk for entanglement which resulted in mortality of two birds. A third bird found entangled was found alive, entangled in vegetation, and was released after I removed the nanotag. I also found that 24 Henslow's Sparrows were able to remove nanotags and several damaged their nanotags. While I do not recommend the use of nanotags on this species in future studies, my study did result in determining fall migratory departure timing of Henslow's Sparrows in Ohio which was previously unknown.

Master of Science, The Ohio State University, 2024, Aerospace Engineering

Various trailing-edge (TE) Coanda active flow control (AFC) actuator configurations were experimentally investigated in a propeller slipstream flow to simulate hover flight. This thesis will provide a comprehensive evaluation of the configuration factors—in terms of spanwise position and geometric parameters—which impact the aerodynamic and control authority performance of TE Coanda AFC actuators in propeller-driven flow with no freestream velocity. Two main types of TE Coanda AFC configuration experiments were conducted. The first type of testing involved varying the spanwise placement and size of TE AFC while maintaining constant internal geometry and circular Coanda profile shape with continuous slot blowing. The spanwise location testing's objectives were to determine the optimal Coanda AFC actuator location relative to the propeller slipstream and compare Coanda flow control effectiveness to that of a traditional deflection control surface. Two trailing-edge Coanda actuator sections of fixed spanwise length were designed, fabricated, and evaluated in terms of lift force and pitching moment generation at varying spanwise locations. Velocity profile measurements for this study for the case of no freestream flow indicated that the propeller slipstream is asymmetric over the NACA 0012 wing and contracts toward the side of the wing on which the propeller blade descends during rotation, where propeller downwash is experienced. This asymmetry indicated that there may be an optimum location for Coanda AFC actuators at the wing trailing edge which couple with momentum from the propeller in regions of peak slipstream velocity. The second type of testing involved varying TE AFC nozzle and surface profiles while maintaining constant spanwise location. These nozzles included both continuous and discrete slot blowing as well as sweeping jets. Surfaces investigated included circular, elliptical, and biconvex profiles. Each configuration was mounted to the trailing edge of (open full item for complete abstract)

Doctor of Philosophy (Ph.D.), University of Dayton, 2023, Electro-Optics

This dissertation is concerned with two distinct applications of volume gratings recorded in photorefractive electro-optic crystals. The first of two applications involves the use of these volume gratings to non-mechanically steer laser beams. A geometric and physical-optics based analysis shows the potential for writing programmable volume gratings in lithium niobate using visible wavelengths in the transmission geometry, and subsequently probing those gratings using infrared wavelengths in a reflection geometry. By appropriate adjustments made to the writing beams, it is shown that both the grating spacing and grating tilt angle can be controlled such that the grating becomes a rotatable Bragg mirror for the incident probe beam, thus steering it to desired angles. The second application of these volume gratings is in image processing. System transfer functions determining the spatial evolution of the reference (input wave) and signal (diffracted wave) beams as they propagate inside a self-pumped volume reflection grating are derived and solved numerically. The solutions are then used to highlight the spatial filtering properties of self-pumped volume reflection gratings, with the focus being on the transmitted (un-diffracted) portion of the reference beam, which is shown to be high-pass spatially filtered. The high-pass spatial filtering manifests as programmable 2-dimensional edge enhancement in the transmitted reference beam. Contrast analysis is done for edge enhanced images, both through simulations and experiments, which show a direct proportionality between the strength of edge enhancement seen in the filtered images and the intensity of the writing beam used to record the grating.

Master of Science in Mathematics, Youngstown State University, 2023, Department of Mathematics and Statistics

A graph G is said to arrow a graph H if every red-blue edge coloring of G presents a monochromatic H, and is written G→H. The down-arrow Ramsey set reports all subgraphs H of a graph G for which G→H. Formally, the down-arrow Ramsey set is a graph G is ↓G:= {H⊆G: G→H }. Calculating this set by way of scientific computing is computationally prohibitive with the resources commonly available to graph theorists and other academics. Using existing research into complete graphs, the down-arrow Ramsey sets for small complete graphs (Kn for 2 ≤ n ≤ 7) can be generated quickly. For larger complete graphs (Kn for 8 ≤ n ≤ 11) specific pre-processing steps are leveraged to speed up calculations in addition to existing data sets. Presented is work on the development of a Python script to generate the down-arrow Ramsey set of a graph through efficient memory management and parallel computing methodologies. The down-arrow generator is used to report new results on complete graphs as well as complete bipartite graphs, and assorted other graphs.

Doctor of Philosophy (Ph.D.), Bowling Green State University, 2023, Mathematics

Staic first introduced the exterior GSC-operad ΛS2Vd as a generalization of the exterior algebra in Staic (2020), which he used to give a linear map detS2:V2⊗6 → k with the property that detS2(⊗1≤i Committee: Mihai Staic Ph.D. (Committee Chair); Xiangdong Xie Ph.D. (Committee Member); Ben Ward Ph.D. (Committee Member); Marlise Lonn Ph.D., N.C.C. (Committee Member) Subjects: Mathematics

Master of Science (M.S.), University of Dayton, 2022, Electro-Optics

Traditional computer vision (CV) approaches are the norm when attempting to extract edge information from an imaged object. These digital approaches are almost always per- formed on 2-D intensity imagery and at times can be computationally expensive depending on the algorithm. The advent of holography enabled imaging beyond 2-D by obtaining 3-D information on the target. This is done by photographically storing the interference pattern of an object field that is interfered with by a reference wave. The spatial frequency spectrum of the hologram can be accessed by taking the holograms Fourier transform. By manipulation of the spatial frequency spectrum, which is also needed for isolating the real or virtual image in off-axis digital holography, edge information can be extracted by high pass spatial filtering the pertinent cropped and centered spectrum. We use simple simulations utilizing 2-D and 3-D objects to show edge enhancement qualities using this approach, and compare its performance to conventional CV techniques, specifically Canny edge detection. Unique filters are created and applied to the cropped and centered spectrum of a holograms Fourier transform before the reconstruction process. This eliminates the need for post-processing of recorded imagery in regards to edge detection, meaning digital holography can be used for 3-D imaging and image processing at the same time.

MARCH, University of Cincinnati, 2022, Design, Architecture, Art and Planning: Architecture

The driving force that governs our lives, and therefore our stories, is time. The linearity of time provides structure to our lives, as does its perpetual motion. One of the few forces that can seemingly counteract the passage of time is space. Space exists outside of time and allows us to inhabit the same spaces as those who lived hundreds of years before us, as well as those who will follow us. Space creates shared experiences between disparate lifetimes. One of the ways that we connect space and time is through storytelling. This dissonance across space and time creates a state of liminality at their intersection. Starting with the myth of the crossroads, this paper explores liminality in the city squares of Savannah, Georgia. As with many old American cities, ghost stories are exceedingly popular, furthering the connection of time through space. After a discussion of a few of the particular stories of Savannah, this paper investigates ways to occupy and activate the space of the crossroads through the lens of theater design. Theaters curate a narrative experience for those that move through them. The stage itself sits at the convergence of two major pathways, that of the actors and that of the patrons. The productions that happen on the stage inhabit the false reality of suspended disbelief. Like the crossroads, the stage is a place where dissonance is allowed and the rules of the natural world bend. A theater, therefore, provides the space necessary to break from our linear timelines through storytelling. The culmination of this document (and subsequent project) is the design of a new community based theater for Savannah, Georgia. The design of this theater aims to provide a platform at the crossroads where the people of Savannah can come together to tell their own stories.

MS, University of Cincinnati, 2022, Engineering and Applied Science: Civil Engineering

All-bolted single-angle connections are simple (shear) connections commonly used in steel structures; however, there is limited guidance on appropriately designing single-angle connections for fire safety in steel structures. In the context of fire, the strength of single-angle connections and their deformation capacity has not been extensively researched and understood. To address these gaps in knowledge, a finite element model was generated and a parametric study performed to evaluate all-bolted single-angle connection behavior under fire. A finite element model was first developed and compared to experimental data at both ambient and elevated temperatures. Due to a lack of detailed single-angle experimental results, the benchmarked model was a double-angle connection. A single-angle connection was then created using similar methods and procedures. Finally, a comprehensive investigation was conducted to examine the various factors that may influence the fire behavior of single-angle connection assemblies. These include: single-angle location, edge distance, gap distance, and load ratio. The base model used for single-angle connections was based on standard building construction practices in the United States. According to the findings of this study, the gap distance and single-angle location concerning the beam neutral axis are the primary factors influencing the behavior of single-angle connections at ambient temperatures. In the case of fire, the principal causes influencing the behavior of single-angle connections are gap distance, edge distance, and load ratio. The controlling modes of failure are bolt to the beam fracture, bolt to the column fracture, tear-out of beam web, and angle.

Doctor of Philosophy, The Ohio State University, 2022, Computer Science and Engineering

Fully autonomous aerial systems (FAAS) combine edge and cloud hardware with UAVs and considerable software support to create self-governing systems. FAAS complete complicated missions with no human piloting by sensing and responding to their environment in real-time. FAAS require highly complex designs to function properly, including layers of on-board, edge, and cloud hardware and software. FAAS also necessitate complex software used for controlling low-level UAV actions, data collection and management, image processing, machine learning, mission planning, and high-level decision-making which must integrate across the compute hierarchy effectively to meet autonomy goals in real-time. The complexity of even a relatively simple FAAS makes efficiency difficult to guarantee. Efficiency, however, is paramount to the effectiveness of a FAAS. FAAS perform missions in resource-scarce environments like natural disaster areas, crop fields, and remote infrastructure installations. These areas have limited access to computational resources, network connectivity, and power. Furthermore, UAV battery lives are short, with flight times rarely exceeding 30 minutes. If FAAS are inefficiently designed, UAV may waste precious battery life awaiting further instructions from remote compute resources, delaying or precluding mission completion. For this reason, it is imperative that FAAS designers carefully choose or design edge hardware configurations, machine learning models, autonomy policies, and deployment models. FAAS have the capability to revolutionize a number of industries, but much research must be done to facilitate their usability and effectiveness. In this dissertation, I outline my efforts toward designing and implementing FAAS that are efficient and effective. This dissertation will focus on the following five topics encompassing design, implementation, and applications of FAAS: §1. Creation of new general and domain-specific machine learning algorithms a (open full item for complete abstract)

Doctor of Philosophy, The Ohio State University, 2022, Computer Science and Engineering

Deep neural networks have not only evolved to produce state-of-the-art artificial intelligence (AI) models, but also generalize well, especially when the network has a large capacity and access to a comparable massive training dataset. These models are computationally and environmentally inefficient. Such research works are known as Red AI. However, it is challenging to develop algorithms that generate cost-effective models for Green AI without sacrificing predictive strength or generality. This dissertation shows how we can effectively integrate domain knowledge, either implicitly or explicitly, into an end-to-end learning pipeline to reduce the training and testing overhead of a neural network. In particular, we exploit the target data and task information to make two types of modifications in the learning pipeline, each of which helps to lower the neural network's run-time costs in unique ways. We re-define the learning task as one or more "simpler" subtask(s) so that the subtask(s) requires fewer parameters and training data. Explicit inclusion of a regularization term in the objective function allows us to limit the exploration space to those that are both relevant and plausible for a given application. We demonstrate the benefits of the above methodologies with the help of three applications with disparate challenges: 1) acoustic event detection, 2) radar classification, and 3) scientific problems based on eigenvalue solvers. Acoustic and radar applications require on-device intelligence and rely on Cortex-M7/M3 microcontroller-based edge devices with limited hardware resources and a low power budget. On the other hand, eigenvalue problems yield memory- and compute-intensive models primarily due to the size of the output layer that grows exponentially with the number of particles in the system. For the acoustic event detector, we exploit unlabeled data from the target domain to constrain knowledge transfer from a large "teacher" model to a smaller "stud (open full item for complete abstract)