Dada a grande quantidade de dados gerados atualmente, apenas uma pequena porção dos mesmos pode ser rotulada manualmente por especialistas humanos. Isso é um desafio comum para aplicações de aprendizagem de máquina. Aprendizado semi-supervisionado aborda este problema através da manipulação dos dados não rotulados juntamente aos dados rotulados. Entretanto, se apenas uma quantidade limitada de exemplos rotulados está disponível, o desempenho da tarefa de aprendizagem de máquina (e.g., classificação) pode ser não satisfatória. Diversas soluções abordam este problema através do uso de uma ensemble de classificadores, visto que essa abordagem aumenta a diversidade dos classificadores. Algoritmos como o co-training e o tri-training utilizam múltiplas partições de dados ou múltiplos algoritmos de aprendizado para melhorar a qualidade da classificação de instâncias não rotuladas através de concordância por maioria simples. Além disso, existem abordagens que estendem esta ideia e adotam processos de votação menos triviais para definir os rótulos, como eleição por maioria ponderada, por exemplo. Contudo, estas soluções requerem que os rótulos possuam um certo nível de confiança para serem utilizados no treinamento. Consequentemente, nem toda a informação disponível é utilizada. Por exemplo: informações associadas a níveis de confiança baixos são totalmente ignoradas. Este trabalho propõe uma abordagem chamada social-training, que utiliza toda a informação disponível na tarefa de aprendizado semi-supervisionado. Para isto, múltiplos classificadores heterogêneos são treinados com os dados rotulados e geram diversas classificações para as mesmas instâncias não rotuladas. O social-training, então, agrega estes resultados em um único rótulo por meio de funções de escolha social que trabalham com agregação de rankings sobre as instâncias. Especificamente, a solução trabalha com casos de classificação binária. Os resultados mostram que trabalhar com o ranking completo, ou seja, rotular todas as instâncias não rotuladas, é capaz de reduzir o erro de classificação para alguns conjuntos de dados da base da UCI utilizados. / Given the huge quantity of data currently being generated, just a small portion of it can be manually labeled by human experts. This is a challenge for machine learning applications. Semi-supervised learning addresses this problem by handling unlabeled data alongside labeled ones. However, if only a limited quantity of labeled examples is available, the performance of the machine learning task (e.g., classification) can be very unsatisfactory. Many solutions address this issue by using a classifier ensemble because this increases diversity. Algorithms such as co-training and tri-training use multiple views or multiple learning algorithms in order to improve the classification of unlabeled instances through simple majority agreement. Also, there are approaches that extend this idea and adopt less trivial voting processes to define the labels, like weighted majority voting. Nevertheless, these solutions require some confidence level on the label in order to use it for training. Hence, not all information is used, i.e., information associated with low confidence level is disregarded completely. An approach called social-training is proposed, which uses all information available in the semi-supervised learning task. For this, multiple heterogeneous classifiers are trained with the labeled data and generate diverse classifications for the same unlabeled instances. Social-training then aggregates these results into a single label by means of social choice functions that work with rank aggregation over the instances. The solution addresses binary classification cases. The results show that working with the full ranking, i.e., labeling all unlabeled instances, is able to reduce the classification error for some UCI data sets used.

Doctorat en sciences sociales, politiques et économiques / info:eu-repo/semantics/nonPublished

Doctorat en Sciences / info:eu-repo/semantics/nonPublished

Défini comme un ensemble de mesures d'intérêt général compatibles avec un fonctionnement concurrentiel des marchés, le service universel est l'obligation pour un opérateur (ou un groupe d'opérateurs) de fournir un service de base d'une qualité donnée à tous les consommateurs à un tarif abordable. Le plus souvent, une contrainte de prix uniforme est également imposée. Jusqu'ici, la littérature économique s'est surtout articulée autour des questions allocatives. Néanmoins, l’argument de la Pareto-efficacité n’est pas suffisant pour justifier pleinement le service universel. En particulier, si la littérature souligne également l'aspect éthique des OSU, aucune référence n'est faite aux développements récents des théories économiques de la justice.Nous proposons d’articuler cette thèse en trois points. Premièrement, nous proposons l’étude des soubassements éthiques des OSU, tels qu’analysés dans la littérature. Cela consiste à proposer une lecture de l’économie industrielle des OSU en termes d’équité et de justice entre les agents impliqués dans les transactions. La notion d’égalitarisme spécifique à la Tobin est traditionnellement avancée pour justifier les OSU. Les nouvelles approches économiques de l’équité peuvent permettre d’aller plus loin et d’affiner les principes en mettant l’accent sur des arbitrages tels que égalité-responsabilité ou égalité-liberté de choix. Dans un deuxième temps, les préférences éthiques du régulateur sont représentées par une fonction de bien-être social à la Atkinson-Stiglitz. Ceci permet d’élucider l’implémentation de critères éthiques à travers les OSU dans un contexte ou il existe des contraintes de marché (telles qu’une tarification ou une couverture de monopole ou de duopole). Ceci révèle les avantages comparatifs de chaque instrument des OSU en termes d’efficacité et d’équité. Cela permet également de déterminer quels instruments sont complémentaires ou substituts au regard des différentes mesures d’équité et d’efficacité. Enfin, nous décrivons l’évolution du cadre législatif du service universel, avant d’exposant l’impact distributif du changement de structure de marché dans les industries de réseaux. Alors que le service universel est souvent présenté comme le parent pauvre de la concurrence, visant à considérer les question de justice sociale, nous montrons que l’introduction de la concurrence s’inscrit en elle-même dans une logique distributive qui lui est propre. / Defined as a set of public interest measures compatible with a competitive environment, universal service is the obligation for an operator (or a group of operators) to provide a range of basic services of specified quality to all consumers at an affordable rate. In many instances, a uniform price is imposed as an additional requirement to the service provider.Until now, the economic literature has focused essentially on efficiency issues. Nevertheless, this Pareto efficiency argument is not sufficient to justify plainly the universal service. Particularly, the literature stresses on the ethical aspect of USO but without using the recent development in distributive justice and without introducing it in a competitive market benchmark.The purpose of this thesis is to analyze the nature and the justification of USO and its relationship with ethic considerations. Why should USO constraints be imposed? What should a regulator do with social preferences in term of coverage, pricing, and market structure? Indeed, most models on coverage and pricing of USO generally assume their implementation exogenously, but do not give a justification for them.We propose to articulate this thesis around three points. First, we propose the study of the underlying ethical principles of USO, as analyzed in the literature, in order to provide a reading of the industrial organization of USO in terms of equity and justice between the agents involved in transactions. The notion of Tobin specific egalitarianism is traditionaly given to justify the USO. The new economic approaches to equity can allow going further and refining the egalitarian principles with emphasis on trade-off such as equality-responsibility and equality-freedom of choice. Secondly, equity preferences of the regulator are represented by an Atkinson-Stiglitz social welfare function. This allows us to study how to implement equity criteria through USO in a context where there is market constraints (such as monopoly or duopolistic pricing and coverage). This reveal the comparative advantages of each USO instruments in terms of efficiency and equity. This also determines which instruments are substitutes or complements with respect to different equity and efficiency measures.Thirdly, we describe the evolution of the legislative framework of universal service. Then we expose the distributional impacts of market structure changes in network industries. While universal service is often presented as the residual of competition intended to consider the issues of social justice, we show that the introduction of competition is rather fit itself in a distributive logic.

Thesis advisor: Uzi Segal / This dissertation consists of three chapters analyzing preferences for randomization in social choice problems. The first two chapters are related and in the fields of distributive justice and social choice. They concern allocation of an indivisible good in social choice problems where efficiency is at odds with equality. The last chapter addresses a social choice problem from an individual's perspective using decision theoretical analysis. In this dissertation I demonstrate why randomization may be an attractive policy in social choice problems and demonstrate how individuals may have preferences over the precise method of randomization. The first chapter is titled "Live and Let Die." This paper discusses how to allocate an indivisible good by social lottery when agents have asymmetric claims. Intuition suggests that there may exist agents who should receive zero probability in the optimal social lottery. In such a case, I say that these agents have weak claims to the good. This paper uses a running example of allocating an indivisible medical treatment to individuals with different survival rates and reactions to the treatment in order to provide conditions for consistency of weak claims. As such, I develop two related assumptions on a social planner's preferences over lotteries. The first -- survival rate scaling -- states that if an individual has a weak claim, then his claim is also weak when survival rates increase proportionally. The second -- independence of weak claims -- states that if an individual has a weak claim, then his removal does not affect others' probabilities of receiving the treatment. These assumptions imply that a compatible social welfare function must exhibit constant elasticity of substitution, which results in potentially-degenerate weighted lotteries. The second chapter is titled "Why is Six Afraid of Seven? Bringing the "Numbers" to Economics." This chapter discusses the numbers problem: the question of if the numbers of people involved should be used to determine whether to help certain people or to help certain other people. I discuss the main solutions that have been proposed: flipping a coin, saving the greater number, and proportionally weighted lotteries. Using the economic tools of social choice, I then show how the model of the previous chapter, "Live and Let Die," can be extended to address numbers problems and compare the implications of prominent social welfare functions for numbers problems. I argue that potentially-degenerate weighted lotteries can assuage the main concerns discussed in the literature and I show that both the Nash product social welfare function as well as constant elasticity of substitution (CES) social welfare functions are compatible with this solution. Finally, I discuss a related problem known as "probability cases," in which individuals differ in survival chances rather than numbers of individuals at risk. When the model is extended to allow for both asymmetries in survival chances and numbers of individuals in groups, CES results in potentially-degenerate weighted lotteries whereas Nash product does not. The third chapter is titled "All Probabilities are Equal, but Some Probabilities are More Equal than Others," which is joint work with Professor Uzi Segal of the Economics Department at Boston College and Professor Shlomo Naeh of the Departments of Talmud and Jewish Thought at The Hebrew University of Jerusalem. In this chapter we compare preferences for different procedures of selecting people randomly. A common procedure for selecting people is to have them draw balls from an urn in turn. Modern and ancient stories (for example, by Graham Greene and the Talmud) suggest that such a lottery may not be viewed by the individuals as "fair.'' In this paper, we compare this procedure with several alternatives. These procedures give all individuals equal chance of being selected, but have different structures. We analyze these procedures as multi-stage lotteries. In line with previous literature, our analysis is based on the observation that multi-stage lotteries are not considered indifferent to their probabilistic one-stage representations. As such, we use a non-expected utility model to understand the preferences of risk-averse individuals over these procedures and show that they may be not indifferent between them. / Thesis (PhD) — Boston College, 2020. / Submitted to: Boston College. Graduate School of Arts and Sciences. / Discipline: Economics.

This thesis presents three chapters. In Chapter 1, I propose a simple one-to-one matching model, where individuals on one side have private information that affects the preferences of the individuals on the other side. I show the existence of the stable and strategy-proof mechanism in this environment. I present an algorithm that defines this mechanism - the Serial Dictatorship algorithm with cutoffs. I also consider the concept of obvious strategy-proofness. I first consider the case where only preferences, but not experience levels, are sellers' private information. For this case, Serial Dictatorship with cutoffs elicits preferences in an obviously strategy-proof way. On the other hand, when only experience levels, but not preferences are private information, I show that there is no obviously strategy-proof and stable mechanism. A consequence of the latter result is that obvious strategy-proofness is incompatible with stability. Chapter 2 considers settings with rich private information - an agent's type may include private information other than just his preferences. In such settings, I identify a necessary condition for obviously strategy-proof implementation of social choice rules. I consider applications to strict preferences, matching and object allocation. The main assumption behind the obvious dominance is that agents might be cognitively limited and can not engage in contingent reasoning at all. This assumption is unreasonably weak compared to the standard assumption that agents can perfectly distinguish contingencies. In Chapter 3, I strengthen it slightly by assuming that agents are able to do at least some contingent reasoning. I define what it means for the strategy to be obviously dominant with respect to a partition of the state space. I call such strategies partition dominant strategies. A strategy is an almost obviously dominant if, for all possible partitions, but not for the coarsest, it can be identified as being partition dominant. My hypothesis is that even though some agents can not do state-by-state reasoning as rational players do, they are able to do at least some partitioning of the other player’s actions and regardless of how the partitioning is done, the agents can identify an almost obviously dominant strategy.

Thesis advisor: Uzi Segal / The dissertation studies the property of transitivity in the social choice theory. I explain why we should care about transitivity in decision theory. I propose two social decision theories: redistribution regret and ranking regret, study their properties of transitivity, and discuss the possibility to find a best choice for the social planner. Additionally, in the joint work, we propose a general method to construct a consistent estimator given two parametric models, one of which could be incorrectly specified. In “Why Transitivity”, to explain behaviors violating transitivity, e.g., preference reversals, some models, like regret theory, salience theory were developed. However, these models naturally violate transitivity, which may not lead to a best choice for the decision maker. This paper discusses the consequences and the possible extensions to deal with it. In “Redistribution Regret and Transitivity”, a social planner wants to allocate resources, e.g., the government allocates fiscal revenue or parents distribute toys to children. The social planner cares about individuals' feelings, which depend both on their assigned resources, and on the alternatives they might have been assigned. As a result, there could be intransitive cycles. This paper shows that the preference orders are generally non-transitive but there are two exceptions: fixed total resource and one extremely sensitive individual, or only two individuals with the same non-linear individual regret function. In “Ranking Regret”, a social planner wants to rank people, e.g., assign airline passengers a boarding order. A natural ranking is to order people from most to least sensitive to their rank. But people's feelings can depend both on their assigned rank, and on the alternatives they might have been assigned. As a result, there may be no best ranking, due to intransitive cycles. This paper shows how to tell when a best ranking exists, and that when it exists, it is indeed the natural ranking. When this best does not exist, an alternative second-best group ranking strategy is proposed, which resembles actual airline boarding policies. In “Over-Identified Doubly Robust Identification and Estimation”, joint with Arthur Lewbel and Jinyoung Choi, we consider two parametric models. At least one is correctly specified, but we don't know which. Both models include a common vector of parameters. An estimator for this common parameter vector is called Doubly Robust (DR) if it's consistent no matter which model is correct. We provide a general technique for constructing DR estimators (assuming the models are over identified). Our Over-identified Doubly Robust (ODR) technique is a simple extension of the Generalized Method of Moments. We illustrate our ODR with a variety of models. Our empirical application is instrumental variables estimation, where either one of two instrument vectors might be invalid. / Thesis (PhD) — Boston College, 2021. / Submitted to: Boston College. Graduate School of Arts and Sciences. / Discipline: Economics.

Title: An overview of selected social procedures Author: Eva Schlosáriková Department: Department of Political Science Supervisor: RNDr. Ondrej Majer, CSc. This thesis first defines social procedures. These are studied and analyzed by social soft- ware, an emerging interdisciplinary field. This thesis will focus on a prominent social procedure, the coalition bargaining and coalition formation. Basic definition and overview of the coalition theory are introduced. Last chapter will concentrate on Bram's model of the coalition formation and his definition of the stable coalition. Other models of the coalition bargaining are briefly introduced and compared to the Bram's model. Keywords: coalition bargaining, coalition formation, stability of coalition 1

No description available.

No description available.