The Internet has radically changed every mode of production, distribution, and consumption in modern life, granting global access to information transcending time and space. However, the physical infrastructure that enables the Internet is often overlooked, due to the virtual and intangible nature of Internet connection. This disconnect between the digital and the physical is captured by the popular buzzword - “the cloud” - a metaphor that abstracts away the vast, energy-guzzling tracts of data server farms kept hidden in corporate secrecy. This thesis seeks to make “the cloud” visible and concrete, and in doing so, zooms in from the Internet’s “global village”1 back to the physical realities of Internet infrastructure embedded within local communities. Through an analysis and synthesis of technical papers, corporate marketing, policy reports, and maps, it situates the “global” nature of the cloud within national and local politics and communities. In addition, through a case study of Google’s first data center in The Dalles, Oregon, and its intimate relationship with a community-piloted municipal fiber network - QLife Network - this thesis brings local community needs and negotiations in conversation with the global corporate order of digital technologies. The discussion concludes with a proposed framework for a community-oriented approach to accessing the global Internet, urging for visibility through new media art and community-based Internet infrastructure investment. Recognizing the realities of local within a global Internet makes the first necessary step in bringing the cloud out of its technological black box, and in short, finding the break in the cloud.
The Internet has radically changed every mode of production, distribution, and consumption in modern life, granting global access to information transcending time and space. However, the physical infrastructure that enables the Internet is often overlooked, due to the virtual and intangible nature of Internet connection. This disconnect between the digital and the physical is captured by the popular buzzword - “the cloud” - a metaphor that abstracts away the vast, energy-guzzling tracts of data server farms kept hidden in corporate secrecy. This thesis seeks to make “the cloud” visible and concrete, and in doing so, zooms in from the Internet’s “global village”1 back to the physical realities of Internet infrastructure embedded within local communities. Through an analysis and synthesis of technical papers, corporate marketing, policy reports, and maps, it situates the “global” nature of the cloud within national and local politics and communities. In addition, through a case study of Google’s first data center in The Dalles, Oregon, and its intimate relationship with a community-piloted municipal fiber network - QLife Network - this thesis brings local community needs and negotiations in conversation with the global corporate order of digital technologies. The discussion concludes with a proposed framework for a community-oriented approach to accessing the global Internet, urging for visibility through new media art and community-based Internet infrastructure investment. Recognizing the realities of local within a global Internet makes the first necessary step in bringing the cloud out of its technological black box, and in short, finding the break in the cloud.

The mosquito Aedes albopictus, originating from southeast Asia, has expanded throughout the world during the 20th century, adapting to a variety of different regions and habitats and spreading diseases such as dengue, chikungunya, and Zika. Within these invasive populations little is known about the genetic diversity and population structure, particularly in Central America. Previous studies on the diversity of Ae. albopictus have been performed in countries such as the United States and Malaysia; however, few studies have examined population structure, genetic diversity, and phylogenetic relationship of Ae. albopictus populations in Central America, where dengue and other viruses are actively spreading. Understanding these important genetic markers and relationships will guide future research to allow for more direct targeting of vector control efforts. Ae. albopictus were captured using five different trapping methods in locations covering a wide range of land uses and temperatures across southern Costa Rica. We assessed population genetic structure by sequencing the mitochondrial Cytochrome C Oxidase 1 (CO1) gene. A phylogenetic tree constructed from our Costa Rican sequences and other world population sequences revealed a slight clustering of the Costa Rican population with populations from temperate regions. Examination of our sequences also revealed little population structuring, which indicates that the geographically separated populations are still intermixing despite the barriers that exist. Understanding the genetics of the Ae. albopictus within Costa Rica will be fundamental to future research to determine how the species will respond to the anthropogenic changes in temperature and land use in southern in Costa Rica. This in turn could aid in proactive mosquito control and public health management to reduce the incidence of the mosquito-borne viral infections.
The mosquito Aedes albopictus, originating from southeast Asia, has expanded throughout the world during the 20th century, adapting to a variety of different regions and habitats and spreading diseases such as dengue, chikungunya, and Zika. Within these invasive populations little is known about the genetic diversity and population structure, particularly in Central America. Previous studies on the diversity of Ae. albopictus have been performed in countries such as the United States and Malaysia; however, few studies have examined population structure, genetic diversity, and phylogenetic relationship of Ae. albopictus populations in Central America, where dengue and other viruses are actively spreading. Understanding these important genetic markers and relationships will guide future research to allow for more direct targeting of vector control efforts. Ae. albopictus were captured using five different trapping methods in locations covering a wide range of land uses and temperatures across southern Costa Rica. We assessed population genetic structure by sequencing the mitochondrial Cytochrome C Oxidase 1 (CO1) gene. A phylogenetic tree constructed from our Costa Rican sequences and other world population sequences revealed a slight clustering of the Costa Rican population with populations from temperate regions. Examination of our sequences also revealed little population structuring, which indicates that the geographically separated populations are still intermixing despite the barriers that exist. Understanding the genetics of the Ae. albopictus within Costa Rica will be fundamental to future research to determine how the species will respond to the anthropogenic changes in temperature and land use in southern in Costa Rica. This in turn could aid in proactive mosquito control and public health management to reduce the incidence of the mosquito-borne viral infections.

In order to survive, organisms must integrate internal and external signals via complex signaling pathways to intelligently navigate their environment. Bacteriophage lambda has long been a classic model for understanding such pathways, and how they allow cells to effect changes in physiology in order to adapt to varying circumstances. When infecting its host Escherichia coli, phage lambda may either make more viruses (lysis), or integrate itself into the host genome (lysogeny). To better understand the extent to which host physiology influences the lysis-lysogeny decision, I performed a forward genetic screen to identify genes whose deletions bias the infected cell towards lysis or lysogeny. This screen revealed numerous host-encoded genes linked to the lysis-lysogeny decision, including transcription factors and genes in metabolic pathways. These results demonstrate previously unknown links between host physiology and viral decision-making, shedding new light on this classic model system.
In order to survive, organisms must integrate internal and external signals via complex signaling pathways to intelligently navigate their environment. Bacteriophage lambda has long been a classic model for understanding such pathways, and how they allow cells to effect changes in physiology in order to adapt to varying circumstances. When infecting its host Escherichia coli, phage lambda may either make more viruses (lysis), or integrate itself into the host genome (lysogeny). To better understand the extent to which host physiology influences the lysis-lysogeny decision, I performed a forward genetic screen to identify genes whose deletions bias the infected cell towards lysis or lysogeny. This screen revealed numerous host-encoded genes linked to the lysis-lysogeny decision, including transcription factors and genes in metabolic pathways. These results demonstrate previously unknown links between host physiology and viral decision-making, shedding new light on this classic model system.

Despite the introduction of Seguro Popular, a universal form of health insurance, in the early 2000s, indigenous women in Mexico continue to exhibit disparately poor maternal health outcomes relative to non-indigenous women. This difference has typically been attributed to factors including poverty, rurality, and lack of care adherence, but little work has addressed the role of prenatal care quality in explaining this trend. Similarly, though Seguro Popular has assisted in equalizing the proportion of indigenous and non-indigenous women who receive prenatal care, few studies have compared the substantive quality of care provided to these two groups. Using data from ENADID 2014, this study characterizes the relationship between indigenous status, insurance affiliation, and prenatal care quality, focusing specifically on the state of Oaxaca. After developing an index of prenatal care quality consistent with both international norms and local patient care preferences, I demonstrate significantly worse prenatal care quality among indigenous women across every measure of quality. Even after controls for socioeconomic marginalization, linear and logistic regression models of prenatal care demonstrate that indigenous women received 0.642 fewer prenatal visits (p < 0.05), exhibited reduced odds of receiving prenatal care in the first trimester (OR 0.5929, p < 0.05), received 0.3515 fewer interventions over the course of pregnancy (out of a checklist of 11 essential procedures, p < 0.05), had reduced odds of receiving prenatal care from a doctor (OR 0.3969, p < 0.01), received poorer information quality of care (0.2284 points fewer on a 3-point scale, p < 0.05), and received poorer interpersonal quality of care (0.3791 points fewer on a 3-point scale, p < 0.01). Regression models stratified by indigenous status suggest that these indigenous status- based inequalities are concentrated among Seguro Popular affiliates. The signifiant correlation between indigenous status and poor care quality, independent of socioeconomic marginalization, suggests insufficient cross-cultural communication and implicit discrimination as potential sources of this quality difference. In addition to demonstrating continued indigenous care access inequality in Oaxaca even after the implementation of Seguro Popular, these results suggest discrepancies in access to quality prenatal care as a possible explanation of Mexico’s indigenous maternal health disparities.
Despite the introduction of Seguro Popular, a universal form of health insurance, in the early 2000s, indigenous women in Mexico continue to exhibit disparately poor maternal health outcomes relative to non-indigenous women. This difference has typically been attributed to factors including poverty, rurality, and lack of care adherence, but little work has addressed the role of prenatal care quality in explaining this trend. Similarly, though Seguro Popular has assisted in equalizing the proportion of indigenous and non-indigenous women who receive prenatal care, few studies have compared the substantive quality of care provided to these two groups. Using data from ENADID 2014, this study characterizes the relationship between indigenous status, insurance affiliation, and prenatal care quality, focusing specifically on the state of Oaxaca. After developing an index of prenatal care quality consistent with both international norms and local patient care preferences, I demonstrate significantly worse prenatal care quality among indigenous women across every measure of quality. Even after controls for socioeconomic marginalization, linear and logistic regression models of prenatal care demonstrate that indigenous women received 0.642 fewer prenatal visits (p < 0.05), exhibited reduced odds of receiving prenatal care in the first trimester (OR 0.5929, p < 0.05), received 0.3515 fewer interventions over the course of pregnancy (out of a checklist of 11 essential procedures, p < 0.05), had reduced odds of receiving prenatal care from a doctor (OR 0.3969, p < 0.01), received poorer information quality of care (0.2284 points fewer on a 3-point scale, p < 0.05), and received poorer interpersonal quality of care (0.3791 points fewer on a 3-point scale, p < 0.01). Regression models stratified by indigenous status suggest that these indigenous status- based inequalities are concentrated among Seguro Popular affiliates. The signifiant correlation between indigenous status and poor care quality, independent of socioeconomic marginalization, suggests insufficient cross-cultural communication and implicit discrimination as potential sources of this quality difference. In addition to demonstrating continued indigenous care access inequality in Oaxaca even after the implementation of Seguro Popular, these results suggest discrepancies in access to quality prenatal care as a possible explanation of Mexico’s indigenous maternal health disparities.

Natural gas consumption has increased in recent decades due to low prices and emissions benefits over coal. The greenhouse gas (GHG) benefits of natural gas over coal require a low upstream emissions profile, in particular with low fugitive emissions of methane. Furthermore, natural gas is unlike oil in that it is highly transport-constrained. Liquefied natural gas (LNG) allows for overseas shipping but comes at significant economic and energetic costs. We worked with a Canadian liquids-rich gas producer to better understand upstream fugitive emissions and assess their efficacy of leak detection and repair (LDAR) programs. We model emissions from their operations and perform life-cycle assessment (LCA) of a hypothetical scenario where they produce 1 billion cubic feet per day of LNG in coastal British Columbia, for consumption in Shenzhen, China. We determine the life- cycle GHG and criteria air pollutant emissions associated with such a project. We find that the LDAR surveys have resulted in decreased number of emissions points and decreased site-wide emissions. Leaks that were fixed from LDAR surveys tended to remain fixed and did not reappear. Likewise, leaks that were not fixed tended to persist and did not go away on their own, indicating that leak persistence is very high. Consequently, LDAR surveys are resulting in emissions reductions, as long as detected leaks are fixed. Repeat LDAR surveys regularly found new emission sources, even without significant site changes occurring, supporting the idea that LDAR surveys must be done regularly to find and fix new emission sources that arise from equipment failure or breakdown. We find that Canada-to-China LNG will result in fewer life-cycle GHG emissions than the same power generated using coal in Asian markets. Studies have estimated Chinese coal power emissions to be anywhere from 868 to 975 g CO2e/kWh, nearly double our results of 408.2 to 547.9 g CO2e/kWh. Our results are lower than prior studies due to low upstream emissions and more efficient LNG production assumptions. LCA studies of LNG have focused on upstream and liquefaction stages, but our work makes it clear that total emissions are dominated by end-use emissions. When considering the climate benefits and drawbacks of LNG, it is critical to understand how the gas will be used.
Natural gas consumption has increased in recent decades due to low prices and emissions benefits over coal. The greenhouse gas (GHG) benefits of natural gas over coal require a low upstream emissions profile, in particular with low fugitive emissions of methane. Furthermore, natural gas is unlike oil in that it is highly transport-constrained. Liquefied natural gas (LNG) allows for overseas shipping but comes at significant economic and energetic costs. We worked with a Canadian liquids-rich gas producer to better understand upstream fugitive emissions and assess their efficacy of leak detection and repair (LDAR) programs. We model emissions from their operations and perform life-cycle assessment (LCA) of a hypothetical scenario where they produce 1 billion cubic feet per day of LNG in coastal British Columbia, for consumption in Shenzhen, China. We determine the life- cycle GHG and criteria air pollutant emissions associated with such a project. We find that the LDAR surveys have resulted in decreased number of emissions points and decreased site-wide emissions. Leaks that were fixed from LDAR surveys tended to remain fixed and did not reappear. Likewise, leaks that were not fixed tended to persist and did not go away on their own, indicating that leak persistence is very high. Consequently, LDAR surveys are resulting in emissions reductions, as long as detected leaks are fixed. Repeat LDAR surveys regularly found new emission sources, even without significant site changes occurring, supporting the idea that LDAR surveys must be done regularly to find and fix new emission sources that arise from equipment failure or breakdown. We find that Canada-to-China LNG will result in fewer life-cycle GHG emissions than the same power generated using coal in Asian markets. Studies have estimated Chinese coal power emissions to be anywhere from 868 to 975 g CO2e/kWh, nearly double our results of 408.2 to 547.9 g CO2e/kWh. Our results are lower than prior studies due to low upstream emissions and more efficient LNG production assumptions. LCA studies of LNG have focused on upstream and liquefaction stages, but our work makes it clear that total emissions are dominated by end-use emissions. When considering the climate benefits and drawbacks of LNG, it is critical to understand how the gas will be used.

Counterinsurgency cultural training (COIN-CT) could be augmented by adapting existing 3D desktop educational modules to virtual reality (VR). The following literature review synthesizes research on VR as a medium for training, VR as a medium for perspective taking, and current COIN-CT offerings to promote future empirical studies on VR COIN-CT solutions. This paper examines how VR features such as social presence, body transfer, and embodied cognition can lead to improved learning transfer. This paper also reviews current COIN-CT offerings in depth, identifies how VR could enhance existing options, and addresses potential challenges facing widespread adoption of VR COIN-CT.
Counterinsurgency cultural training (COIN-CT) could be augmented by adapting existing 3D desktop educational modules to virtual reality (VR). The following literature review synthesizes research on VR as a medium for training, VR as a medium for perspective taking, and current COIN-CT offerings to promote future empirical studies on VR COIN-CT solutions. This paper examines how VR features such as social presence, body transfer, and embodied cognition can lead to improved learning transfer. This paper also reviews current COIN-CT offerings in depth, identifies how VR could enhance existing options, and addresses potential challenges facing widespread adoption of VR COIN-CT.

This study looked at personalization during the 2016 presidential election. The ubiquity of social media and the unique features of Twitter enable new forms of personalization to emerge during political campaigns. This research added to the growing body of literature in political communication by taking a multifaceted approach to personalized politics in assessing content, interactivity, and language. The findings in this study confirmed trends from previous research in that politicians are posting more content relating to traditional campaign strategies (agenda setting, mobilization, and criticism) rather than revealing personal information about themselves. However, findings also suggest that personalized communicative practices are indeed emerging on Twitter through use of interactive features and language. This multifaceted analysis allows one to draw inferences to how personalization takes form, to give insights on how political campaigning will continue to evolve on Twitter, and to discuss its implications on the democratic process.
This study looked at personalization during the 2016 presidential election. The ubiquity of social media and the unique features of Twitter enable new forms of personalization to emerge during political campaigns. This research added to the growing body of literature in political communication by taking a multifaceted approach to personalized politics in assessing content, interactivity, and language. The findings in this study confirmed trends from previous research in that politicians are posting more content relating to traditional campaign strategies (agenda setting, mobilization, and criticism) rather than revealing personal information about themselves. However, findings also suggest that personalized communicative practices are indeed emerging on Twitter through use of interactive features and language. This multifaceted analysis allows one to draw inferences to how personalization takes form, to give insights on how political campaigning will continue to evolve on Twitter, and to discuss its implications on the democratic process.

An Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) reveals information about open chromatin regions, individual nucleosomes, and chromatin compaction at nucleotide resolution using only 500 to 50,000 cells. In contrast, Chromatin Immunoprecipitation sequencing (ChIP-seq) requires much more biological samples, typically millions of cells, to detect DNA regions with histone modifications. In this sense, regressing histone ChIP-seq data from less costly ATAC-seq data will help us map missing histone marks and understand epigenomic activity in a more efficient way. This paper investigates how our modified deep adversarial training approach can be used to predict ChIP-seq signal based on ATAC-seq signal. We begin by setting the performance of convolutional neural network (CNN) model as a baseline. We then introduce three modifications to the widely used adversarial network architecture. First, we modify the generator component of the adversarial network so that it takes ATAC-seq signal as input instead of random noise and generates ChIP-seq signal from the ATAC-seq signal. Second, we suggest composite objective function based on two different losses - mean squared error and adversarial loss. Third, we apply one-sided label smoothing, which is essential in stabilizing the adversarial training. The generator trained through our new adversarial training approach reports Pearson correlation of 0.562 with respect to the actual ChIP-seq signal, outperforming the CNN baseline. We also conduct qualitative analysis on how the adversarial training based on the composite objective function helps the model predict ChIP-seq peaks using ATAC-seq signal. To the best of our knowledge, this is the first attempt to tackle epigenomic signal imputation task using deep adversarial training.
An Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) reveals information about open chromatin regions, individual nucleosomes, and chromatin compaction at nucleotide resolution using only 500 to 50,000 cells. In contrast, Chromatin Immunoprecipitation sequencing (ChIP-seq) requires much more biological samples, typically millions of cells, to detect DNA regions with histone modifications. In this sense, regressing histone ChIP-seq data from less costly ATAC-seq data will help us map missing histone marks and understand epigenomic activity in a more efficient way. This paper investigates how our modified deep adversarial training approach can be used to predict ChIP-seq signal based on ATAC-seq signal. We begin by setting the performance of convolutional neural network (CNN) model as a baseline. We then introduce three modifications to the widely used adversarial network architecture. First, we modify the generator component of the adversarial network so that it takes ATAC-seq signal as input instead of random noise and generates ChIP-seq signal from the ATAC-seq signal. Second, we suggest composite objective function based on two different losses - mean squared error and adversarial loss. Third, we apply one-sided label smoothing, which is essential in stabilizing the adversarial training. The generator trained through our new adversarial training approach reports Pearson correlation of 0.562 with respect to the actual ChIP-seq signal, outperforming the CNN baseline. We also conduct qualitative analysis on how the adversarial training based on the composite objective function helps the model predict ChIP-seq peaks using ATAC-seq signal. To the best of our knowledge, this is the first attempt to tackle epigenomic signal imputation task using deep adversarial training.

This thesis presents the design methodology of a miniaturized, portable, and low power electrostatic precipitator (ESP) to reduce indoor air pollution (IAP) from rural cook-stoves. Decreasing the concentration of aerosol particles helps reduce the incidence of respiratory tract infections that can lead to disease and even death. We trace existing technologies and standards to combat IAP in cookstoves, design an ESP that allows precipitation at lower voltages via an electrode design that utilizes sub 5 kV voltages, and explore how rapidly pulsed converters decrease the power consumption of the circuit and increase smoke collection.
This thesis presents the design methodology of a miniaturized, portable, and low power electrostatic precipitator (ESP) to reduce indoor air pollution (IAP) from rural cook-stoves. Decreasing the concentration of aerosol particles helps reduce the incidence of respiratory tract infections that can lead to disease and even death. We trace existing technologies and standards to combat IAP in cookstoves, design an ESP that allows precipitation at lower voltages via an electrode design that utilizes sub 5 kV voltages, and explore how rapidly pulsed converters decrease the power consumption of the circuit and increase smoke collection.

While great progress has been made in image classification using machine learning, often achieving near-human or even superhuman accuracy on image classification tasks, recent studies have found that image classification models are vulnerable to adversarial attacks: special images crafted to fool the models into mislabeling the picture. In this work, we investigate the problem of creating an adversarially robust feature: a feature f whose value at any point x cannot be changed much by perturbing x slightly. We establish strong connections between adversarially robust features and a natural spectral property of the geometry of the dataset and metric of interest. This connection can be leveraged both to provide robust features and to provide a lower bound on the robustness of any function that has significant variance across the dataset. Finally, we provide empirical evidence that the adversarially robust features yielded via this spectral approach can be fruitfully leveraged to learn a robust (and accurate) model.
While great progress has been made in image classification using machine learning, often achieving near-human or even superhuman accuracy on image classification tasks, recent studies have found that image classification models are vulnerable to adversarial attacks: special images crafted to fool the models into mislabeling the picture. In this work, we investigate the problem of creating an adversarially robust feature: a feature f whose value at any point x cannot be changed much by perturbing x slightly. We establish strong connections between adversarially robust features and a natural spectral property of the geometry of the dataset and metric of interest. This connection can be leveraged both to provide robust features and to provide a lower bound on the robustness of any function that has significant variance across the dataset. Finally, we provide empirical evidence that the adversarially robust features yielded via this spectral approach can be fruitfully leveraged to learn a robust (and accurate) model.

Since 2003, the Sudanese government and its Arab militia proxies have conducted a genocide against Africans in Darfur, killing roughly 500,000 people and displacing 2.7 million. Public outcry compelled President George W. Bush to take diplomatic measures to end the atrocities, though they achieved little success. In 2007, then-Senator Barack Obama included Darfur on his presidential platform and denounced the genocide as a “stain on our souls.” Nevertheless, during his presidency, Obama not only failed to address the genocide, but also lifted sanctions that had been imposed on Khartoum for its criminal activities in Darfur. This thesis examines why the Obama Administration’s foreign policies in relation to Darfur were so weak. In particular, I focus on four main factors that constrained U.S. action in the region: lack of political salience, South Sudan's creation and subsequent implosion, humanitarian crises elsewhere, and China's and Russia's economic interests in Sudan. By understanding these policy constraints, we might learn how to end this long and violent genocide, as well as other humanitarian conflicts in areas with little national strategic import.
Since 2003, the Sudanese government and its Arab militia proxies have conducted a genocide against Africans in Darfur, killing roughly 500,000 people and displacing 2.7 million. Public outcry compelled President George W. Bush to take diplomatic measures to end the atrocities, though they achieved little success. In 2007, then-Senator Barack Obama included Darfur on his presidential platform and denounced the genocide as a “stain on our souls.” Nevertheless, during his presidency, Obama not only failed to address the genocide, but also lifted sanctions that had been imposed on Khartoum for its criminal activities in Darfur. This thesis examines why the Obama Administration’s foreign policies in relation to Darfur were so weak. In particular, I focus on four main factors that constrained U.S. action in the region: lack of political salience, South Sudan's creation and subsequent implosion, humanitarian crises elsewhere, and China's and Russia's economic interests in Sudan. By understanding these policy constraints, we might learn how to end this long and violent genocide, as well as other humanitarian conflicts in areas with little national strategic import.

This dramaturgical thesis interrogates the transnational dimension of staging Frances Ya-Chu Cowhig's play 'The World of Extreme Happiness,' a play about a factory girl in Shenzhen China, at Stanford University, an elite neoliberal university down the road from Silicon Valley. Specifically, we argue that, as Western cosmopolitan subjects, the audience encounter with the play is that of cultural consumption, projecting a fantasized timeline of modernity onto the protagonist's journey, with her rural home seen as 'backwards' and 'local' and the Western city as 'progressive' and 'cosmopolitan.' Consequently, this thesis examines how a production of this play can disrupt such an imagination by invoking the ways the protagonist is 'modern' and cosmopolitan, while still confronting the transnational power dynamics present in the relationship between audience and protagonist. We propose theatrical devices, particularly in and around 'gazing,' to performatively subvert these imagined global hierarchies among a Western cosmopolitan audience.
This dramaturgical thesis interrogates the transnational dimension of staging Frances Ya-Chu Cowhig's play 'The World of Extreme Happiness,' a play about a factory girl in Shenzhen China, at Stanford University, an elite neoliberal university down the road from Silicon Valley. Specifically, we argue that, as Western cosmopolitan subjects, the audience encounter with the play is that of cultural consumption, projecting a fantasized timeline of modernity onto the protagonist's journey, with her rural home seen as 'backwards' and 'local' and the Western city as 'progressive' and 'cosmopolitan.' Consequently, this thesis examines how a production of this play can disrupt such an imagination by invoking the ways the protagonist is 'modern' and cosmopolitan, while still confronting the transnational power dynamics present in the relationship between audience and protagonist. We propose theatrical devices, particularly in and around 'gazing,' to performatively subvert these imagined global hierarchies among a Western cosmopolitan audience.