Strategically using information to affect the views of a population is not a new phenomenon and dates all the way back to the earliest development of political systems. Some of the earliest examples still in existence today were produced by a Greek writer Herodotus who penned an inaccurate account of historical events in the Persian empire approximately 500 years after the birth of Jesus Christ. Political propaganda is nearly as old as politics. Similarly, fake news, as a specific tactic of propaganda, is not relegated only to the information age. Older examples are easy to find that pre-date the digital revolution. However, the speed of distribution and the number of people that can be reached by leveraging the modern information infrastructure is unprecedented. These factors combine to result in increased risk from fake news that must be addressed. The internet has enabled the distribution of vast amounts of information to an incredibly large population virtually instantaneously and for comparatively low cost. While the development of this capability has resulted in enormous economic development and provided great benefit to the world, it has also exposed the same population to increased risk. The rapid distribution of fake news can cause contagion, manipulate markets, spark conflict, or fracture strategic relationships. The scourge of fake news is even more problematic with the current limitations on fact-checking methodologies which are unable to keep pace with the increased volume of fake news production. In the search for methods to combat fake news, both public and private organizations are struggling. Probabilistic Risk Analysis (PRA) can be leveraged to quantitatively describe the risk associated with fake news. This thesis presents a method for modeling management decisions designed to combat fake news in an online network. It leverages established infectious disease modeling to describe online virality and implements countermeasure regimes to inform opposition decision making. The model was informed by two online surveys of a representative sample of the U.S. voting population that endeavored to measure the impact of limited but targeted fake news. The results point to both the potential effectiveness and limitations of fake news that is leveraged as part of a targeted influence campaign. The online survey results also point to the dangers of the use of modified video and audio, known as "deep fakes, " in fake news of the future. Technological improvements including expertly crafted deep fakes, online microtargeting, smart trolls, and the potential use of artificial intelligence for content production suggests fake news will continue to persist as a scourge in the future and could present a viable threat to democratic self-governance

Community-based innovation has recently emerged as a focal point of research on strategy and innovation. Scholars have argued that communities have several advantages over firms, such that managers must increasingly consider how to best compete and collaborate with this unique organizational form. Extant empirical work provides significant insight. However, it remains unclear how and when firms benefit from user community-based innovation. Across three tightly-linked papers, this dissertation addresses this gap. The first is a comprehensive review of prior work that studies user communities as they relate to firm strategy and performance. The second paper is a comparative case study of innovation by two civilian drone ventures—one that organized as a community, the other that organized as a firm. Finally, using a novel panel dataset of 2,586 video game development projects, the third paper develops and tests theory on whether firms can improve performance via learning from communities. Together, these papers contribute several interrelated findings to literature on firm strategy, organization theory and user community innovation, as well as several insights for management practice.

This dissertation utilizes a unique setting—coding bootcamps—to examine how workers attempt to reskill themselves for growing job areas without traditional organizations serving as the backdrop for their actions. I draw on 80 semi-structured interviews and observations conducted at two bootcamps in Silicon Valley over the course of 17 months of fieldwork. My findings suggest that bootcamps resembled learning collectives where self-learning and learning from peers and near-peers figured more prominently than expert instruction. Under conditions of minimal expert instruction and obstacles to legitimate peripheral participation, I show how aspiring software developers sought out an occupational community in virtual spaces, learned asynchronously from unknown others, developed their practice through mock work among themselves and managed to get hired as a new category of occupational entrant—the bootcamp graduate. This dissertation contributes to our understanding of employability management practices and under-institutionalized learning and socialization processes in contemporary careers.

This research presents a warning systems model in which early-stage cyber threat signals are generated using machine learning and artificial intelligence (AI) techniques. Cybersecurity is most often, in practice, reactive. Based on the manual forensics of machine-generated data by humans, security efforts only begin after a loss has taken place. The current security paradigm can be significantly improved. Cyber-threat behaviors can be modeled as a set of discrete, observable steps called a 'kill chain.' Data produced from observing early kill chain steps can support the automation of manual defensive responses before an attack causes losses. However, early AI-based approaches to cybersecurity have been sensitive to exploitation and overly burdensome false positive rates resulting in low adoption and low trust from human experts. To address the problem, this research presents a collaborative decision paradigm with machines making low-impact/high-confidence decisions based on human risk preferences and uncertainty thresholds. Human experts only evaluate signals generated by the AI when decisions exceed these thresholds. This approach unifies core concepts from the disciplines of decision analysis and machine learning by creating a super-agent. An early warning system using these techniques has the potential to avoid more severe downstream consequences by disrupting threats at the beginning of the kill chain.

This dissertation examines the potential for electric vehicles to act as a viable technology pathway to deep decarbonization of the transportation sector. Taking an integrative and multi-disciplinary approach, this research applies theories of technology development and diffusion, social welfare optimization, life cycle analysis, and empirical modeling to better understand the processes and prospects for widespread adoption, as well as the emissions benefits that could accrue from a full market transition. The first chapter develops an endogenous model of market diffusion which incorporates positive feedback effects such as learning-by-doing and network externalities, then uses it to consider what an optimal subsidy policy regime could look like in different representative scenarios. The next chapter goes more in-depth on emissions, presenting the first comparative full life cycle analysis of greenhouse gas emissions for mass market, long-range battery electric vehicles, as compared to internal combustion engine vehicles. Lastly, data from real-world trips are utilized to explore heterogeneity in vehicle efficiency and range under different trip conditions, factors which could have significant implications for the scalability of the electric vehicle technology. Broadly, results show very large electric vehicle emissions reductions across all markets, robustness of key performance metrics to local climate conditions, and positive social welfare impacts of subsidies from accelerating early market adoption.

Forward contracting can deliver benefits in terms of risk hedging, price discovery, and market power mitigation. Aiming to advance the understanding of its role in market design, I examine how the introduction of a forward market changes the strategic and hedging incentives of participants. I challenge the two common literature assumptions of (i) one-sided market power: the assumption that only sellers exercise market power and (ii) perfect arbitrage: the equivalence of forward and (expected) spot price. I provide a framework in which oligopolists face oligopsonists in two sequential markets and allow for a price premium to arise endogenously as the result of hedging and strategic considerations. The introduction of a forward market is found to increase efficiency but does not necessarily do so in a Pareto improving way. The distribution of potential benefits to sellers and buyers depends on their risk aversion and market power. Policy implications are discussed.

Over the past fifty years, financial and legal theory have predicted that owners of firms may have incentives to take risks that have positive expected value from the perspective of firms' equity holders but negative expected value from the perspective of total firm value and of society as a whole. These incentives stem from the fact that a firm's equity holders can enjoy potentially unlimited upside payoffs when risks turn out well, but have capped downside payoffs equal to the value of their equity holdings. Theory also predicts that the lower the value of those equity holdings (relative to total firm assets), the more pronounced these incentive effects may become. This dissertation encompasses three pieces of research that investigate these theoretical matters empirically. The investigations provide empirical evidence amongst banks and amongst industrial firms that as the value of equity (relative to assets) decreases, socially sub-optimal risk-taking increases. The investigations also provide empirical evidence that when a legal change increases the extent to which industrial firms' creditors will bear losses on account of harms that a bankruptcy debtor has caused to third parties, those creditors can exert their influence on borrowers to constrain their socially inefficient risk-taking.

Despite the significant annual expenditure of public (government) research and development (R& D) funding allocated to private, for-profit firms, research is unclear as to if whether or not public funding is positively associated with firm performance. To partially address this gap, in three essays I examine the most common funding tie between public organizations and private firms in the U.S. - public-private R& D relationships. In the first two essays, I conceptualize public-private R& D relationships as institutional hybrids and argue and show that relative to technologies developed without a public-sector partner, the institutional conflict underlying public-private R& D relationships is positively related to the creation of valuable and destabilizing technologies. To provide support for my institutional hybrids' hypotheses, I employ a novel machine learning-matching method to examine all patented technologies in the U.S. between the years 1982-2012. In the final essay, I focus on technology ventures (i.e., start-ups) that form a contractual R& D relationship with a public-consumer (e.g., NASA, DoD) and show that these start-ups produce technologies faster and survive longer than their non-contracted peers. However, due to the stabilizing dependencies inherent in public-consumer relationships, these ventures also experience slower growth. I conclude with management and policy implications.

Artificial Intelligence systems, powered by algorithms, are seemingly everywhere: in corporations, hospitals, classrooms, cars, and even baby monitors. Despite the prevalence and breathless rhetoric, fewer studies have examined how these systems impact people in practice. This dissertation focuses on understanding the impact of algorithms on people using online labor markets to find work. Existing organizational and social science theory suggests that artificial intelligence and algorithmic systems are "tightening" the iron cage by providing a more sophisticated, rational method of controlling the way people work and interact. This dissertation, however, reveals how an online labor market's use of algorithms is creating modern day invisible cages, in which platforms deliberately hide the norms and expectations for how people should behave. To unpack and theorize this central finding, I draw upon field work collected from one of the world's largest online labor markets. Using participant observation, interviews, and extensive archival data, I first show how people using the online labor market unwittingly gave the platform full permission to implement algorithms that were completely opaque to them. I highlight how the way in which people agree to the terms of service in an online labor market context represents a significant shift compared to previous agreements people consented to in similar, offline labor market settings. I then show how the online labor market implemented an inscrutable algorithm dictating people's success on the platform. The implementation of this algorithm led to what I call "superstitious reactivity, " because people were never able to verify if and how their actions contributed to their success on the online labor market. Taken together, this dissertation has implications for our understanding and the literature on algorithms, evaluations, reactivity, reputation, labor market intermediaries, employment relationships, digital contracts, and agnotology.

Technological innovations are fundamentally transforming information flow and social behavior. Digital platforms are increasingly embedded in established social systems highlighting the importance of developing fair and private machine learning approaches to managing these platforms. As social networks and social media data are increasingly leveraged for business and policy decisions, this thesis discusses work to develop methods for analyzing data privacy in attribute-rich network data, and to promote fair and equitable digital systems. Chapter 2 discusses work on privacy in digital systems. Privacy attacks in networked systems commonly leverage homophilous interactions, or what's commonly referred to as "birds of a feather flock together". While homophily describes a bias in attribute preferences for similar others, it gives limited attention to variability. We observe that attribute preferences can exhibit variation beyond what can be explained by models of homophily, and observe that this excess variation can induce a similarity among friends-of-friends on a network without requiring any similarity among friends. These findings offer an alternative perspective on network structure and attributes in general and prediction in particular, complicating the already difficult task of protecting privacy on social networks. Then, Chapter 3 examines node attribute prediction tasks more generally and provides a framework for distinguishing different types of prediction tasks as within-network, across-network, or across-layer tasks. This work highlights that methods aimed at across-network tasks are in fact evaluated on across-layer problems and have limited performance on across-network problems. Chapter 4 assesses novel uses of social media data to target regulatory enforcement efforts. This work provides a more cautionary perspective on the leading case that Yelp reviews are useful to target food safety inspections. This work highlights that prior results are sensitive to ``extreme imbalanced sampling'': extreme because the dataset was restricted from roughly 13k inspections to a sample of only 612 inspections with only extremely high or low inspection scores, and imbalanced by not accounting for class imbalance in the population. Finally, Chapter 5 studies the challenge of class imbalance more generally, comparing more standard approaches like Synthetic Minority Oversampling Technique to more modern approaches like Weighted Random Forest and Balanced Random Forest. Taken together, this thesis highlights the importance of understanding how new technologies intersect with existing consumer behavior and demonstrate the unique role of machine learning and social science research in the design and management of data products. As technology continues to promote instantaneous connectivity, there is a pressing need for analytical tools that facilitate secure and equal-opportunity platforms.

The thesis contains three parts which provide for an in depth study of problems that are motivated from challenges faced by online platforms. The first two parts study matching problems in decentralized and centralized environments respectively and the third part is concerned with the aggregation of preferences. The first part of the thesis considers the problem of assortment planning in two-sided platforms, where self-interested agents are present on each side of the platform. While assortment planning has been studied extensively in the revenue management literature, it is concerned with the problem of assigning menus of goods (as opposed to people who can reject) to agents. In our problem, agents' utilities are derived from a random utility choice model (RUM). Two different design settings are explored - simultaneous (both sides propose at the same time) and sequential (one side proposes first). The optimal menu is completely characterized for the case of a single customer where a greedy algorithm is shown be optimal. Both design settings turn out to be NP-Hard for multiple customers. Two heuristics are proposed and a constant factor approximation algorithm is provided for the sequential setting. The second part of the thesis considers a centralized marketplace, where agents (drivers and passengers) arrive in an online manner and are matched after a possible delay. Agents arrive at a given set of locations which form a metric (distance between points satisfying the triangle inequality) and the platform needs to minimize the sum of distance and waiting costs. A natural trade-off that arises is the following: matching agents faster reduces waiting costs but waiting for more agents to arrive allows to create better matches. A randomized and a deterministic algorithm are proposed on the tree metric (that has been embedded from the original metric); both of which achieve a constant competitive ratio on the tree metric leading to a logarithmic competitive ratio on the general metric. The third part of the thesis looks at pairwise comparisons among agents specifically in the context of intransitivity. The goal is to develop and analyze parametric models that can support intransitivity. The key result is that such models have non-concave likelihood functions. Next, taking inspiration from the Condorcet paradox, a majority vote model is proposed. The majority vote model can support strong and arbitrarily long intransitive cycles and shows comparable performance against other parametric models which support intransitivity on publicly available datasets.

As datasets capturing human choices and rankings grow in richness and scale, particularly in online domains, there is an increasing need for more flexible models of ranking and choice. I introduce the Pairwise Choice Markov Chain (PCMC) model of discrete choice, which escapes traditional choice-theoretic axioms such as regularity and stochastic transitivity while remaining inferentially tractable. PCMC still satisfies the axiom of uniform expansion, a considerably weaker assumption than Luce's choice axiom, and PCMC retains the Multinomial Logit (MNL) as a special case. Using the notion of a choice representation to break a ranking into a sequence of choices, I develop a framework for turning models for discrete choice into models for ranking data, generalizing the idea used to build the seminal Plackett-Luce (PL) model from MNL. This framework allows us to leverage our progress in choice into novel models for ranking data and allows us to build flexible yet tractable models for complete and partial rankings from recent advances in choice models. I also discuss methods and subtleties of shrinkage estimation for choice data and demonstrate the efficacy of the models I propose in prediction tasks from a broad range of domains including transportation, competitions, voting, and sushi.

A majority of infectious disease models are compartmental models, due to their simplicity and interpretability. However the assumption of homogeneous mixing in compartmental models does not hold in general, and for certain diseases or subpopulations the heterogeneity in mixing behavior is particularly prominent. With more and more data becoming available to shed light on the heterogeneous mixing of population, network model has emerged as an alternative solution. This dissertation focuses on developing and applying network models to study infectious diseases with readily available data. In Chapter 2, we propose a mathematical framework to study the spread of blood-borne diseases among people who inject drugs (PWID). The prevalence rate of blood-borne diseases is higher among PWID as compared to the general population, due to unprotected sex and needle-sharing. To accurately account for transmissions via both types of risky contact, we build a bi-layer network model. We present methodology for inferring model parameters from readily available data sources. The model can be used to evaluate interventions that exploit the structure of the contact network to contain infectious diseases in PWID. To illustrate, we instantiate a network model with data collected by a needle and syringe program in Chicago, then we use the model to evaluate the potential effects of a peer education (PE) program on averting new HIV and HCV infections over the next ten years. We show that a targeted PE program would avert significantly more HIV and HCV infections than an untargeted program, highlighting the importance of reaching individuals who are centrally located in contact networks. In Chapter 3, we adapt the network model of Chapter 2 to evaluate the cost-effectiveness of a pre-exposure prophylaxis (PrEP) program for PWID. Inspired by the illustrative example of targeted PE program in Chapter 2, we sought to improve the cost-effectiveness of PrEP program by enrolling high risk PWID. Specifically we investigate four strategies: 1) random PWID are enrolled (Unselected Enrollment); 2) individuals are randomly selected and enrolled together with their partners (Enroll Partners); 3) individuals with the highest number of sexual and needle-sharing partnerships are enrolled (Most Partners); 4) individuals with the greatest number of infected partners are enrolled (Most Positive Partners). We measure costs and QALYs associated with each strategy, and find that all strategies except the Unselected Enrollment strategy are cost-effective, indicating that selection of high risk PWID for PrEP can indeed improve the cost-effectiveness of PrEP for PWID. In Chapter 4, we explore a household model to study the circulation of poliovirus in a community.The Polio Eradication and Endgame Strategic Plan 2013-2018 calls for the gradual withdrawal of OPV from routine immunization. Little is understood regarding the transmission potential of Sabin virus when it is reintroduced in the post-cessation era. In view of this, we develop an individual-based stochastic epidemic model and use the model to investigate two scenarios: 1) an outbreak response using OPV takes place to interrupt wild poliovirus transmission; 2) a case actively shedding Sabin virus is released into the community and sparks an undetected circulation. We find that inactivated poliovirus vaccine (IPV) alone is insufficient to prevent long term circulation of Sabin virus under current level of sanitation and contact patterns in Mexican communities if OPV is reintroduced. Our result highlights the need for aggressive OPV use in outbreak response, vigilant poliovirus surveillance, improvement in hygiene and sanitation, as well as development of new vaccines that protect against fecal shedding and reinfection. We conclude with a summary of our findings and a discussion of future research directions in Chapter 5.

Convex optimization is the cornerstone of continuous optimization, but many real problems arising in machine learning and operations research are nonconvex. This two part thesis explores my work developing principled algorithms for finding local minima of nonconvex functions. Part I: The complexity of finding stationary points of nonconvex functions. For a long time, it was known that gradient descent achieved an $\epsilon^{-2}$ rate for finding stationary points of unconstrained nonconvex functions, but better rates for first-order methods were unknown. We show that, using an algorithm that judiciously utilizes Nesterov's accelerated gradient descent, it is possible to improve this rate to $\epsilon^{-7/4}$ rate for functions with Lipschitz first and second derivatives. Adding Lipschitz third derivatives improves this rate to $\epsilon^{-5/3}$. Moreover, we provide almost matching lower bounds to prove that (i) finding a stationary point is easier for convex functions and (ii) acceleration in nonconvex optimization requires assumptions beyond smoothness. Part II: Interior point methods for nonconvex optimization. The second part of this thesis focuses on an interior point method (IPM) for optimization problems with nonconvex constraints. The work of Wachter and Biegler suggests that infeasible-start interior point methods (IPMs) developed for linear programming cannot be adapted to nonlinear optimization without significant modification, i.e., using a two-phase or penalty method. We propose an IPM that, by careful initialization and updates of the slack variables, is guaranteed to find a first-order certificate of local infeasibility, local optimality or unboundedness of the (shifted) feasible region. Our proposed algorithm differs from other IPM methods for nonconvex programming because we reduce primal feasibility at the same rate as the barrier parameter. This gives an algorithm with more robust convergence properties and closely resembles successful algorithms from linear programming. Comparisons with IPOPT on a subset of CUTEst problems indicate less frequent failures and superior performance for detecting infeasibility. Finally, we develop a highly simplified version of our IPM method with nonconvex constraints that obtains an $\epsilon$-scaled KKT point in $\epsilon^{-7/4}$ iterations. This provides the first proof that IPMs with nonconvex constraints have a polynomial runtime in $1/\epsilon$.

As machine learning systems increasingly get applied in high-stake domains such as autonomous vehicles and medical diagnosis, it is imperative that they maintain good performance when deployed. Modeling assumptions rarely hold due to noisy inputs, shifts in environment, unmeasured confounders, and even adversarial attacks to the system. The standard machine learning paradigm that optimize average performance is brittle to even small amounts of noise, and exhibit poor performance on underrepresented minority groups. We study \emph{distributionally robust} learning procedures that explicitly protect against potential shifts in the data-generating distribution. Instead of doing well just on average, distributionally robust methods learn models that can do well on a range of scenarios that are different to the training distribution. In the first part of thesis, we show that robustness to small perturbations in the data allows better generalization by optimally trading between approximation and estimation error. We show that robust solutions provide asymptotically exact confidence intervals and finite-sample guarantees for stochastic optimization problems. In the second part of the thesis, we focus on notions of distributional robustness that correspond to uniform performance across different subpopulations. We build procedures that balance tail-performance alongside classical notions of average performance. To trade these multiple goals \emph{optimally}, we show fundamental trade-offs (lower bounds), and develop efficient procedures that achieve these limits (upper bounds). Then, we extend our formulation to study partial covariate shifts, where we are interested in marginal distributional shifts on a subset of the feature vector. We provide convex procedures for these robust formulations, and characterize their non-asymptotic convergence properties. In the final part of the thesis, we develop and analyze distributionally robust approaches using Wasserstein distances, which allows models to generalize to distributions that have different support than the training distribution. We show that for smooth neural networks, our robust procedure guarantees performance under imperceptible adversarial perturbations. Extending such notions to protect against distributions defined on learned feature spaces, we show these models can also improve performance across unseen domains.

The problem that motivates this dissertation is why some decision problems have automated support tools, while others remain manual. Economic theory suggests that for any good or service where demand increases, an entrepreneur will appear to help fill the gap between supply and demand. The framework of decision analysis has shown a normative approach to the decision-making process, and since its inception, many have tried to explain the technical requirements needed to achieve intelligent decision systems or automated decision support. But the market has thus far failed to leverage this body of knowledge to create automated decision support business at scale. This dissertation presents an analytic framework for the creation of automated decision businesses and allows entrepreneurs who wish to enter this market to have a quantitative approach to understand which barriers have the most significant effect on the lag of automated decision support. This theory provides us with the ability to measure the impact of implementing new ideas and how future research could affect those barriers for any particular decision problem. One of the main barriers to automation is the need for personalization that is common in most decision problems. This barrier stems from the difficulty of eliciting preferences from decision-makers and encoding the elicitation process in a way that lends itself to automation. This dissertation provides a mathematical approach to preference elicitation (Preference Thresholds) that significantly improves the ability of decision-makers to interact with automated decision systems as well as reducing the implementation costs of automation, and lowering the cognitive burden on untrained decision-makers. Preference Thresholds allow for the piecewise discretization of continuous value functions around points of interest (preference) to each decision-maker. PTs also represent a way to reduce the size of uncertainties around preferences and to frame problems to eliminate alternatives that do not fit in with the decision maker's values.\ Another critical barrier to automation is the difficulty of implementing these models at scale. This dissertation provides an example of how broad model templates can be used to reduce the difficulty by providing willing entrepreneurs with a starting point for a particular type of decision. In the application chapter, this dissertation introduces a broad model template for patient-centric medical decision making and proposes a way to customize that template to many conditions as well as a way to automate the customization process. The template, together with preference thresholds (PT), is used to create, develop, and deploy an automated decision support tool to aid with the treatment selection decision of prostate cancer patients. The tool is implemented at Stanford Cancer Center for patients with clinically localized prostate cancer who come for a consult with a radiation oncologist.

Decision makers must often choose a course of action under limited time with limited knowledge. In this work, we formalize longstanding observations about the efficacy of improper linear models to construct accurate yet easily applied rules that can help resource-constrained practitioners make better informed decisions that are consistent with their stated objectives. We find that simple rules can help substantially improve the performance of human experts, while rivaling the accuracy of complex prediction models that base decisions on considerably more information. Policy makers, however, may be reluctant to adopt such analytical tools due to the difficulty in anticipating, prior to deployment, the impact of resulting policies. In particular, one generally cannot use historical data to directly observe what would have happened had the recommended actions been taken. To address this issue, we present two strategies for gauging the sensitivity of predicted policy outcomes to potentially unmeasured confounders. We further investigate how humans compare to systematic rules when explicitly asked to assess risk of recidivism in the context of criminal justice. Our results show that systematic rules often perform better than humans in terms of both accuracy and ranking. Finally, we introduce a simple three-step strategy---which we call risk-adjusted regression---for measuring disparities that existing policies may have across different groups defined by race, gender, and other protected attributes that may be of interest. One popular strategy is to estimate disparities after controlling for observed covariates, typically with a regression model. This approach, however, suffers from two statistical challenges: omitted- and included-variable bias. Our proposed method addresses both concerns in settings where decision makers have clearly measurable objectives.

This thesis covers several applications of Operations Research in the domains of finance and healthcare. There are three chapters, each covering a different application. Chapter 1 applies techniques from deep learning to estimate mortgage risk. The near-elimination of feature engineering is substituted by models with several thousand parameters that require large amounts of training data. The predictive performance of these models strongly exceeds that of baseline models, especially for predicting prepayments. This increased accuracy, however, comes at the cost of a more opaque model which is harder for a human to interpret than simpler models like logistic regression, which are currently the industry standard. The work in Chapter 2 explores the design of policies for biometric authentication. It first develops a model for the joint distribution of similarity scores associated with different fingers and irises. In the second step, this model is harnessed to design near-optimal multi-stage policies that would be used for authentication, and are robust to gaming, can be computed in real-time and are personalized for optimal performance. The work shows that a reduction of several orders of magnitude in the error rates is achievable by solely changing the authentication policies -- and leaving the hardware unchanged. Chapter 3 is motivated by a humanitarian cause geared towards helping developing countries -- to reduce mortality in children by identifying effective interventions at the planning stage. It takes a descriptive health model (called LiST), which estimates mortality of children given coverage of interventions, and embeds that into an optimization engine in order to minimize mortality under a fixed budget. In doing so, it allows LiST to be used in a prescriptive framework, where policymakers can identify the optimal intervention set at a fixed budget as well as recognize the trade-off of mortality reduction and budget allocation. We find that a greedy strategy offers near-optimal performance with ease of implementation. The findings also highlight the critical role that optimization plays in mortality reduction.

Better understanding energy, how to harness and utilise it effectively, and what the implications of its use in these ways means from an individual to a more aggregate level are important topics for business, policy making, and research. Growing evidence pointing to the serious global consequences of society's energy usage further increases the importance of these goals. Furthermore it is increasingly clear that the costs and benefits of energy use are experienced disproportionately across the population. The changing nature of individual decision-making and how the actions of individuals affect an entire economic system also play important roles in our effort to implement effective energy and environmental policy. Both on the demand-side and the supply-side, trends of decentralisation and the increased adoption of more granular technologies are making it increasingly important to understand the role of the individual and their behaviour within the broader energy-economic-environmental system. This dissertation contributes to the intersection of the energy, environmental, and behavioural economics literatures with an emphasis on policy efficacy. It considers questions of efficient retail electricity pricing, externality-correcting policies, the role of behavioural bias in consumer decision making, and the implications of such bias for policy efficacy. Its approach is grounded in neoclassical economics, but it brings the newer field of behavioural economics into the fold where its inclusion strengthens the ability to answer the question at hand. There are two main projects in this dissertation. The first is an in-depth evaluation of the rooftop solar pricing policy of Net Energy Metering (NEM) from both an economic efficiency and distributional equity approach. Treatment of this topic is approached first theoretically and then complemented with a suite of numerical simulations. The second project represents an attempt to codify and unify considerations of behavioural bias in individual decision making and what implications they have for optimal policy efficacy. In this way it showcases behavioural economics as an extension of, rather than a challenge to, neoclassical theory.

Consider a farm of robots operating concurrently, learning to carry out a task, and sharing data with one another in real time. There are clear benefits to scale, since a larger number of robots can gather and share more data in the same amount of time that enables each robot in the group to learn faster than it would learn on its own. However, the benefits of parallelization can be greatly magnified if the robots explore the environment in a coordinated fashion, diversifying their experience and adapting appropriately as data is gathered. We identify three properties which are necessary for efficient coordinated exploration -- adaptivity, commitment, diversity -- and demonstrate that straightforward extensions of statistically efficient optimistic and posterior sampling approaches from single-agent to concurrent reinforcement learning fail to satisfy them. As an alternative, we propose seed sampling, which extends posterior sampling in a manner that meets these requirements. We proceed to design concurrent reinforcement learning algorithms that achieve the three properties of efficient coordinated exploration in real systems with potentially enormous state spaces, to which tabular methods do not scale. This is achieved by coupling the seed sampling concept with learning general value function or policy function representations that allow the agents to generalize and operate in arbitrarily large scale environments. Overall, this dissertation presents a concept that enables rapid learning of effective policies in the presence of complex goals, when we have multiple agents who learn to operate in parallel in a common, unknown environment and collaborate in real time. There is a multitude of applications that can benefit from this algorithmic framework, ranging from recommendation systems, to robot automation, to autonomous vehicles.