Unsupervised image-to-image translation (UNIT) is a challenging task in the image manipulation field, where input images in a visual domain are mapped into another domain with desired visual patterns (also called styles). An ideal direction in this field is to build a model that can map an input image in a domain to multiple target domains and generate diverse outputs in each target domain, which is termed as multi-domain and multi-modal unsupervised image-to-image translation (MMUIT). Recent studies have shown remarkable results in UNIT but they suffer from four main limitations: (1) State-of-the-art UNIT methods are either built from several two-domain mappings that are required to be learned independently or they generate low-diversity results, a phenomenon also known as model collapse. (2) Most of the manipulation is with the assistance of visual maps or digital labels without exploring natural languages, which could be more scalable and flexible in practice. (3) In an MMUIT system, the style latent space is usually disentangled between every two image domains. While interpolations within domains are smooth, interpolations between two different domains often result in unrealistic images with artifacts when interpolating between two randomly sampled style representations from two different domains. Improving the smoothness of the style latent space can lead to gradual interpolations between any two style latent representations even between any two domains. (4) It is expensive to train MMUIT models from scratch at high resolution. Interpreting the latent space of pre-trained unconditional GANs can achieve pretty good image translations, especially high-quality synthesized images (e.g., 1024x1024 resolution). However, few works explore building an MMUIT system with such pre-trained GANs. In this thesis, we focus on these vital issues and propose several techniques for building better MMUIT systems. First, we base on the content-style disentangled framework and propose to fit the style latent space with Gaussian Mixture Models (GMMs). It allows a well-trained network using a shared disentangled style latent space to model multi-domain translations. Meanwhile, we can randomly sample different style representations from a Gaussian component or use a reference image for style transfer. Second, we show how the GMM-modeled latent style space can be combined with a language model (e.g., a simple LSTM network) to manipulate multiple styles by using textual commands. Then, we not only propose easy-to-use constraints to improve the smoothness of the style latent space in MMUIT models, but also design a novel metric to quantitatively evaluate the smoothness of the style latent space. Finally, we build a new model to use pretrained unconditional GANs to do MMUIT tasks.

No description available.

No description available.

No description available.

We developed a deep fusion methodology of non-destructive (NDT) in-situ infrared and ex- situ ultrasonic images for localization of porosity detection without compromising the integrity of printed components that aims to improve the Laser-based additive manufacturing (LBAM) process. A core challenge with LBAM is that lack of fusion between successive layers of printed metal can lead to porosity and abnormalities in the printed component. We developed a sensor fusion U-Net methodology that fills the gap in fusing in-situ thermal images with ex-situ ultrasonic images by employing a U-Net Convolutional Neural Network (CNN) for feature extraction and two-dimensional object localization. We modify the U-Net framework with the inception and LSTM block layers. We validate the models by comparing our single modality models and fusion models with ground truth X-ray computed tomography images. The inception U-Net fusion model localized porosity with the highest mean intersection over union score of 0.557.

Yes / Recent research has shown that adults with autism spectrum disorders (ASD) have difficulty interpreting others' emotional responses, in order to work out what actually happened to them. It is unclear what underlies this difficulty; important cues may be missed from fast paced dynamic stimuli, or spontaneous emotional responses may be too complex for those with ASD to successfully recognise. To explore these possibilities, 17 adolescents and adults with ASD and 17 neurotypical controls viewed 21 videos and pictures of peoples' emotional responses to gifts (chocolate, a handmade novelty or Monopoly money), then inferred what gift the person received and the emotion expressed by the person while eye movements were measured. Participants with ASD were significantly more accurate at distinguishing who received a chocolate or homemade gift from static (compared to dynamic) stimuli, but significantly less accurate when inferring who received Monopoly money from static (compared to dynamic) stimuli. Both groups made similar emotion attributions to each gift in both conditions (positive for chocolate, feigned positive for homemade and confused for Monopoly money). Participants with ASD only made marginally significantly fewer fixations to the eyes of the face, and face of the person than typical controls in both conditions. Results suggest adolescents and adults with ASD can distinguish subtle emotion cues for certain emotions (genuine from feigned positive) when given sufficient processing time, however, dynamic cues are informative for recognising emotion blends (e.g. smiling in confusion). This indicates difficulties processing complex emotion responses in ASD.

A door-to-door trip may involve multiple traffic modes. For example, travelers may drive to a subway station and make a transfer to rail transit; alternatively, people may also start their trips by walking/cycling to a bus/subway station and then take transit in most of the trip. A successful eco-route planning thus should be able to cover multiple traffic modes and offer intermodal routing suggestions. Developing such a system requires to address extensive concerns. The dissertation is a building block of the multi-modal energy-efficient routing system which is being developed and tested in the simulation environment before real applications. Four submodules have been developed in the dissertation as partial fulfillment of the simulation-based system: energy consumption modeling, subway system development, on-road vehicles dynamic eco-routing, and information effect on route choice behavior. Other submodules such as pedestrian/bicycle modeling will be studied in the future. Towards the research goal, the dissertation first develops fuel consumption models for on-road vehicles. Given that gasoline light duty vehicles (LDVs) and electric vehicles were modeled in previous studies, the research effort mainly focuses on heavy duty vehicles (HDVs). Specifically, heavy duty diesel trucks (HDDTs) as well as diesel and hybrid-electric transit buses are modeled. The models are developed based on the Virginia Tech Comprehensive Power-based Fuel consumption Modeling (VT-CPFM) framework. The results demonstrate that the model estimates are highly consistent with field observations as well as the estimates of the Comprehensive Modal Emissions Model (CMEM) and MOtor Vehicle Emissions Simulator (MOVES). It is also found that the optimum fuel economy cruise speed ranges between 32 and 52 km/h for the tested trucks and between 39 and 47 km/h for the tested buses on grades varying from 0% to 8%, which is significantly lower than LDVs (60-80 km/h). The dissertation then models electric train dynamics and energy consumption in support of subway simulation system development and trip energy estimation. The dynamics model varies throttle and brake level with running speed rather than assuming constants as was done by previous studies, and the energy consumption model considers instantaneous energy regeneration. Both models can be easily calibrated using non-engine data and implemented in simulation systems and eco-transit applications. The results of the dynamics modeling demonstrate that the proposed model can adequately capture instantaneous acceleration/deceleration behavior and thus produce realistic train trajectories. The results of the energy consumption modeling demonstrate that the model produces the estimates consistent with the National Transit Database (NTD) results, and is applicable for project-level analysis given its ability in capturing the energy consumption differences associated with train, route and operational characteristics. The most suitable simulation testbed for system development is then identified. The dissertation investigates four state-of-the-art microsimulation models (INTEGRATION, VISSIM, AIMSUM, PARAMICS). Given that the car-following model within a micro-simulator controls longitudinal vehicle motion and thus determines the resulting vehicle trajectories, the research effort mainly focuses on the performance of the built-in car-following models from the energy and environmental perspective. The vehicle specific power (VSP) distributions resulting from each of the car-following models are compared to the field observations. The results demonstrate that the Rakha-Pasumarthy-Adjerid (RPA) model (implemented in the INTEGRATION software) outperforms the Gipps (AIMSUM), Fritzsche (PARAMICS) and Wiedemann (VISSIM) models in generating accurate VSP distributions and fuel consumption and emission estimates. This demonstrates the advantage of the INTEGRATION model over the other three simulation models for energy and environmental analysis. A new eco-routing model, comprehensively considering microscopic characteristics, is then developed, followed by a numerical experiment to test the benefit of the model. With the resulting eco-routing model, an on-road vehicle dynamic eco-routing system is constructed for in-vehicle navigation applications, and tested for different congestion levels. The results of the study demonstrate that the proposed eco-routing model is able to generate reasonable routing suggestions based on real-time information while at the same time differentiate eco-routes between vehicle models. It is also found that the proposed dynamic eco-routing system achieves lower network-wide energy consumption levels compared to the traditional eco-routing and travel time routing at all congestion levels. The results also demonstrate that the conventional fuel savings relative to the travel time routing decrease with the increasing congestion level; however, the electric power savings do not monotonically vary with congestion level. Furthermore, the energy savings relative to the traditional eco-routing are also not monotonically related to congestion level. In addition, network configuration is demonstrated to significantly affect eco-routing benefits. The dissertation finally investigates the potential to influence driver behavior by studying the impact of information on route choice behavior based on a real world experiment. The results of the experiment demonstrate that the effectiveness of information in routing rationality depends upon the traveler's age, preferences, route characteristics, and information type. Specifically, information effect is less evident for elder travelers. Also, the provided information may not be contributing if travelers value other considerations or one route significantly outperforms the others. The results also demonstrate that, when travelers have limited experiences, strict information is more effective than variability information, and that the faster less reliable route is more attractive than the slower more reliable route; yet the difference becomes insignificant with experiences accumulation. The results of the study will be used to enhance system design through considering route choice incentives. / Ph. D.

Climate change impacts hydrology, nutrient cycling, agricultural conservation practices, and greenhouse gas (GHG) emissions. The Chesapeake Bay and its watershed are subject to the largest and most expensive Total Maximum Daily Load (TMDL) ever developed. It is unclear if the TMDL can be met given climate change and variability (e.g., extreme weather events). The objective of this dissertation is to quantify the impact of climate change and climate on water resources, nutrient cycling and export in agroecosystems, and agricultural conservation practices in the Chesapeake Bay watershed. This is accomplished by developing and employing a suite of modelling tools. GHG emissions from agroecosystems, particularly nitrous oxide (N2O), are an increasing concern. To quantify N2O emissions a routine was developed for the Soil and Water Assessment Tool (SWAT) model. The new routine predicts N2O and di-nitrogen (N2) emissions by coupling the C and N cycles with soil moisture, temperature, and pH in SWAT. The model uses reduction functions to predict total denitrification (N2 + N2O production) and partitions N2 from N2O using a ratio method. The SWAT nitrification routine was modified to predict N2O emissions using reduction functions. The new model was tested using GRACEnet data at University Park, Pennsylvania, and West Lafayette, Indiana. Results showed strong correlations between plot measurements of N2O flux and the model predictions for both test sites and suggest that N2O emissions are particularly sensitive to soil pH and soil N, and moderately sensitive to soil temperature/moisture and total soil C levels. The new GHG model was then used to analyze the impact of climate change and extreme weather conditions on the denitrification rate, N2O emissions, and nutrient cycling/export in the 7.4 km2 WE38 watershed in Pennsylvania. Climate change impacts hydrology and nutrient cycling by changing soil moisture, stoichiometric nutrient ratios, and soil temperature, potentially complicating mitigation measures. To quantify the impact of climate change we forced the new GHG model with downscaled and bias-corrected regional climate model output and derived climate anomalies to assess their impact on hydrology, nitrate (NO3-), phosphorus (P), and sediment export, and on emissions of N2O and N2. Model-average (± standard deviation) results indicate that climate change, through an increase in precipitation, will result in moderate increases in winter/spring flow (2.7±10.6 %) and NO3- export (3.0±7.3 %), substantial increases in dissolved P (DP, 8.8±19.8 %), total P (TP, 4.5±11.7 %), and sediment (17.9±14.2 %) export, and greater N2O (63.3±50.8 %) and N2 (17.6±20.7 %) emissions. Conversely, decreases in summer flow (-12.4±26.7 %) and the export of P (-11.4±27.4 %), TP (-7.9±24.5 %), sediment (-4.1±21.4 %), and NO3- (-12.2±31.4 %) are driven by greater evapotranspiration from increasing summer temperatures. Increases in N2O (20.1±29.3 %) and decreases in N2 (-13.0±14.6 %) are also predicted in the summer and driven by increases in soil moisture and temperature. In an effort to assess the impact of climate change at a regional level, the model was then scaled-up to the entire Susquehanna River basin and was used to evaluate if agricultural best management practices (BMPs) can offset the impact of climate change. Agricultural BMPs are increasingly and widely employed to reduce diffuse nutrient pollution. Climate change can complicate the development, implementation, and efficiency of BMPs by altering hydrology, nutrient cycling, and erosion. We select and evaluate four common BMPs (buffer strips, strip crop, no-till, and tile drainage) to test their response to climate change. We force the calibrated model with six downscaled global climate models (GCMs) for a historic period (1990-2014) and two future scenario periods (2041-2065) and (2075-2099) and quantify the impact of climate change on hydrology, NO3-, total N (TN), DP, TP, and sediment export with and without BMPs. We also tested prioritizing BMP installation on the 30% of agricultural lands that generate the most runoff (e.g., critical source areas-CSAs). Compared against the historical baseline and excluding the impact of BMPs, the ensemble model mean (± standard deviation?) predictions indicate that climate change results in annual increases in flow (4.5±7.3%), surface runoff (3.5±6.1%), sediment export (28.5±18.2%) and TN (9.5±5.1%), but decreases in NO3- (12±12.8%), DP (14±11.5%), and TP (2.5±7.4%) export. When agricultural BMPs are simulated most do not appreciably change the overall water balance; however, tile drainage and strip crop decrease surface runoff generation and the export of sediment, DP, and TP, while buffer strips reduced N export substantially. Installing BMPs on critical source areas (CSAs) results in nearly the same level of performance for most practices and most pollutants. These results suggest that climate change will influence the performance of BMPs and that targeting BMPs to CSAs can provide nearly the same level of water quality impact as more widespread adoption. Finally, recognizing that all of these model applications have considerable uncertainty associated with their predictions, we develop and employ a Bayesian multi-model ensemble to evaluate structural model prediction uncertainty. The reliability of watershed models in a management context depends largely on associated uncertainties. Our Objective is to quantify structural uncertainty for predictions of flow, sediment, TN, and TP predictions using three models: the SWAT-Variable Source Area model (SWAT-VSA), the standard SWAT model (SWAT-ST), and the Chesapeake Bay watershed model (CBP-model). We initialize each of the models using weather, soil, and land use data and analyze outputs of flow, sediment, TN, and TP for the Susquehanna River basin at the Conowingo Dam in Conowingo, Maryland. Using these three models we fit Bayesian Generalized Non - Linear Multilevel Models (BGMM) for flow, sediment, TN, and TP and obtain estimated outputs with 95% confidence intervals. We compare the BGMM results against the individual model results and straight model averaging (SMA) results using a split time period analysis (training period and testing period) to assess the BGMM in a predictive fashion. The BGMM provided better predictions of flow, sediment, TN, and TP compared to individual models and the SMA during the training period. However, during the testing period the BGMM was not always the best predictor; in fact, there was no clear best model during the testing period. Perhaps more importantly, the BGMM provides estimates of prediction uncertainty, which can enhance decision making and improve watershed management by providing a risk-based assessment of outcomes. / Ph. D.

The pavement system is subjected to complex stress states under vehicular loading. A combination of axial and shear stress has been identified as a potential cause of top down cracking (or more precisely near surface cracking) in asphalt surface. Therefore, in terms of modeling the material response a pertinent question is whether the typical one-dimensional viscoelastic properties of the material are affected by a multi-axial stress state. Such changes are referred to as interaction non-linearity. The objective of this study was to evaluate whether or not asphalt composites are susceptible to such interaction effects. The study was conducted using fine aggregate matrix (FAM), which comprises graded sand and asphalt binder. To provide multi-modal loading, the rectangular prismatic FAM specimens were used with the Arcan apparatus. This apparatus ensures low bending stress and offers adjustments in the setup to provide different proportions of axial and shear stress. Finite element modeling was done to evaluate the stress state for different orientations of the sample in the Arcan apparatus. For measurement of strain, the study used digital image correlation (DIC), which is an optical, non-contact measurement technology. The strain thus measured was used to compute shear compliance. Fitting parameters of the shear compliances were estimated for power-law and Prony series for different loading orientations. When compared, the measured shear compliances do not show perceivable variation with respect to different proportion of axial stress applied in conjunction. However, further testing with different temperatures and other magnitudes of shear stress is necessary. This study is the first step to allow modeling of stress and crack propagation behavior near the pavement surface where complex stress state is present. / text

In contemporary consumer culture, branding is the term given to the creation of an image or text (visual, aural, textural or multi-sensory) intended to represent a commodity or product sold by a producer or service provider. This product’s commercial viability depends largely on the way it is presented (via branding) to its target market. The aim of this research report is to show that music used consciously as a branding medium, with special reference to film music (in its commodified form), has become a brand in itself, as opposed to merely a component of a multi-modal commercial product. Through analyses of a central film music theme from Star Wars: Episode IV, composed by John Williams, I aim to identify what I will term `audio-branding techniques’ within the music, thereby showing how music has come to be regarded as a brand. The audio branding techniques will relate directly to the four levels of analysis that I propose to conduct. The nature of branding implies the presence of three entities in the cultural and commercial `transaction’ that takes place: namely, the service provider (creator), the product (commodity) and the target market (consumer). I intend to argue that, as a result of powerful creative collaborations between John Williams and his various directors (not to mention his own unique talent), this composer’s film music has increasingly become an audio brand which is almost commensurate with the brand status of the film itself. Williams’s ability to create a symbiotic relationship between a music brand and that of a film has set him apart from most other contemporary art and commercial composers. As a result, it is not simply the actors, directors and producers associated with a movie that induce one to buy tickets to see it, but Williams’s independent audio branding style as well. I thus aim to prove that his film music is an audio brand independent of, and yet also allied with, other brands.