Coalitions are important mobilization tools that can help build powerful movements and help achieve political outcomes. However, coalition formation is not equally feasible within all social movements. Indeed, existing studies show that intergroup antagonisms stemming from identity-based differences, power asymmetries, and legitimacy concerns impede collaboration within structurally divided movements. What conditions facilitate coalition formation within structurally divided movements? To answer this question, this project examines coalition formation in a particularly hard case, namely the Indian farmers movement. The deep-seated caste, class, and ideological divisions have historically impeded coalition formation within the movement. Yet, in the recent past, large, diverse coalitions have emerged despite the persistence of structural divisions. Drawing on extensive fieldwork (200+ interviews and participant observation of 19 key events) and secondary sources, I argue that mobilizing structural networks have a paradoxical effect on inter-organizational collaboration. A high capacity to mobilize structural networks help organizations attain their objectives, while also exacerbating divisions by giving them the ability to threaten each other’s interests and assert their identity. Conversely, a reduction in the capacity to mobilize structural networks creates greater space for alternative strategies and actors and modify inter-organizational dynamics to make structural divisions less pronounced. Specifically, in the Indian case, the declining capacity made the broader movement environment conducive for coalitions by reducing power asymmetries, delegitimizing structural identities as viable frames of mobilization, and creating more space for new organizations that mobilized beyond structural identities and used collaboration from the very outset. This enabled organizations to first engage with each other on ad-hoc basis, which in turn had a reinforcing effect leading to ties among disparate movement actors. These interconnected organizations coalesced into large, diverse coalitions to confront an oppressive and negligent government. Ultimately, this study sheds light on a potential collective action paradox inherent in divided movements. It reveals that factors enhancing the collective action capacity of individual organizations may simultaneously impede collective action between organizations. As exemplified by the Indian case, this study also underscores that coalitions are a powerful mobilization tool that authorities cannot dismiss as routine opposition, especially so for divided groups.

In many engineering and machine learning applications, we often encounter optimization problems (e.g., resource allocation, clustering) for which finding the exact solution is computationally intractable. In such problems, ad-hoc approximate solutions are often used, which have no performance guarantees. Our goal is to develop approximate optimization methods with the following features a) provable performance guarantees, and b) computational tractability. In this dissertation, we focus on several challenging problems in resource allocation and machine learning and develop optimization methods for the same.

In the first part of this dissertation, we develop optimization methods to solve fundamental resource allocation problems encountered in the design of different systems, namely wireless networks, crowdsourcing systems, and healthcare systems.

Dense deployment of heterogeneous small cells (e.g., picocells, femtocells) is becoming the most effective way to combat the exploding demand for the wireless spectrum. Given the large-scale nature of these deployments, developing resource sharing policies using a centralized system can be computationally and communicationally prohibitive. To this end, we propose a general framework for distributed multi-agent resource sharing. We show that the proposed framework significantly outperforms the state-of-the-art. We prove quite general constant factor approximation guarantees with respect to the optimal solutions.

Matching platforms for freelancing (e.g., Upwork) are becoming mainstream. These platforms are faced with the challenging task of allocating workers to clients in order to generate maximum revenues, taking into consideration that both sides are self-interested, have limited information about the other, and desire to be matched with the best possible partners. We propose a dynamic matching mechanism that takes these challenges into account and achieves many of the aforesaid properties.

Screening plans are used for the early detection of several diseases, such as breast cancer and colon cancer. These screening plans are not personalized to the history and demographics of the subject and can often lead to a delay in the detection of the disease and in other cases cause unnecessary invasive tests such as biopsies. We show that constructing exactly optimal personalized screening plans that minimize the number of screens given a tolerance on the delay is computationally intractable. We develop a framework to solve the proposed problem approximately. We establish general performance guarantees and show that the proposed solution is computationally tractable. We apply the framework to breast cancer screening and establish its utility in comparison to the existing clinical guidelines.

In the second part of this dissertation, we develop optimization methods useful for machine learning applications. Machine learning models are increasingly becoming a part of many of the decision making systems, for instance, clinical decision support systems. Many of the machine learning models are hard to interpret and thus are often called "black-box" models. We propose a method that approximates the black-box models using piecewise-linear approximations. This approach helps explain the model using linear models in different regions of the feature space. We provide provable fidelity, i.e., how well does approximation reflect the black-box, guarantees and show that the method is computationally tractable. We carry out experiments on different datasets and establish the utility of our approach.

Kullback-Leibler divergence is a fundamental quantity used in many disciplines, such as machine learning, statistics, and information theory. We develop an optimization-based approach to estimate the Kullback-Leibler divergence, which relies on the Donsker-Varadhan representation. The state-of-the-art estimator based on this representation relies on solving a non-convex optimization problem and hence, is not consistent. We propose a convex reformulation to construct an estimator, which we show is consistent. We also carry out experiments to show that the proposed estimator is better than the competing estimator.

This dissertation considers counter-institutional archives as social and political gathering places and interweaves spatial analyses of the forces of gentrification affecting the surrounding geographic context of these archival spaces. I developed a method called “queer hanging out” which addresses many related activities of knowledge production with varying stakes. Queer hanging out takes place across political affinities, within and while maintaining the movement spaces discussed in the chapters to follow. Chapter One traces three archival projects in Park Slope Brooklyn, New York, Interference Archive, the Lesbian Herstory Archives, and Moncada Library. I analyze the development of these archives as political movement infrastructure resisting processes of displacement. As movement infrastructure I discuss archival preservation strategies, and programming as political education. Chapter Two discusses several projects in Barcelona to consider political labor, common vs. public space, archival histories, and futurity. In this chapter I think about two contemporary art pieces that were exhibited at the Centre d’Art Santa Mónica, alongside a social center in the Sants Neighborhood called Can Battló and two archives housed there, Fundacio Salvador Segui and Centre de Documentació de Moviment Socials Mercè Grenzner. In each of these examples I focus on how forms of political skepticism toward institutions give way to a political orientation called prefigurative politics. In Chapter Three I write about a recently closed counter-institutional archive in San Francisco, the Center for Sex and Culture (CSC). Already concerned over the preceding years about their ability to maintain the physical space in light of rising rents in the surrounding South of Market neighborhood, CSC’s all-volunteer collective spent two years planning the relocation of the archival materials. In the end, instead of dividing the materials according to contemporary categories and redistributing them to other counter-institutions, CSC’s collective decided to relocate the archive as a whole to Harvard’s Schlesinger library. The political decisions CSC members made explores questions around community-based archives and the ways they have become circumscribed under neoliberal capitalism. Counter-institutional spaces offer us the opportunity to land our rage, grief and solidarity among friends and those we share political affinity, and with whom we want to hang out.

Under what conditions do individuals opt out of the political mainstream and vote for the far-right? Results of recent European elections seem to defy the expectations of the literature, as the far-right gained a large share of votes despite relatively “normal” economic conditions and the stability of institutional constraints. In this dissertation, I focus on the demand-side to uncover why some people with anti-immigrant attitudes vote for the far-right but others with similar attitudes do not when supply-side conditions are held constant. I argue that what separates these voters is how immigrants are perceived as incompatible with the host society. Specifically, I argue that individuals who have populist far-right attitudes and who deem immigrants as “other” because they are not integrated are more likely to vote for the far-right than are those who do not. I develop and test this hypothesis using original survey data collected in Germany. The multivariate quantitative analyses show that the far-right AfD vote is associated with the belief that immigrants are not integrated and a lack of confidence in Chancellor Angela Merkel. Analysis of open-ended responses of AfD voters confirms these findings. AfD voters say that they think immigrants are forging “parallel societies” and that they are concerned about the long-term impact of immigrants in Germany. AfD voters place blame on both immigrants themselves, who they perceive as not integrating, as well as on Angela Merkel, who they blame for allowing immigrants into the country. By uncovering the underlying motivations of “less extreme” far-right voters, the findings of this dissertation provide a greater understanding of the mainstreaming of the far-right and its electoral success in western societies with immigrant populations.