Search

Multi-agent systems are an ubiquitous presence in our everyday life: our entire society could be seen as a huge multi-agent system in which each individual has to perform in an environment populated by other entities, each motivated by its own goals and objectives. In a cooperative system, all of these entities act towards a common goal. This setting has gained a lot of popularity in the AI research community, as many real-world situations can naturally be modelled as such. Still, optimally solving such systems remains a challenging problem. Multi-agent reinforcement learning (MARL) is one of the most employed techniques used to tackle these: agents learn how to behave by repeatedly interacting with their own environment. Although major key steps have been taken in this direction, some fundamental issues arising from the presence of multiple agents that learn and act together remain. In this work, two such aspects are addressed: team representation and the multi-agent credit assignment problem. The former is about the way in which a system designer has to represent and learn the team of agents: on one hand, representing the whole team as a single centralized entity may seem compelling, but this solution is not adaptable to scale to larger systems. On the other hand, learning each agent independently from the others solves that issue, but introduces non-stationarity in the agent learning experiences due to the now ignored presence of the other agents. In this work, the focus is on factorization techniques, as a middle ground between these two extremes: this idea has recently gained a major interest and served as the basis for many recent deep MARL algorithms. Although well performing in practice, all of these methods only focused on single-agent decompositions, leaving the idea of "higher-order" factorizations almost unexplored. Moreover, although factorizations are widely considered capable of improving performances over the two approaches detailed above, no wide investigation of the real merits or general applicability of factored techniques has been conducted so far. This work fills the gap by investigating a wide array of factored methods on a diverse set of cooperative scenarios, to assess their performance both in terms of accuracy of the represented functions and action selection. About the multi-agent credit assignment problem instead, many techniques have been proposed, including difference rewards (one of the most popular family of algorithms used to tackle such problems), but few have been extended to the deep MARL framework. One of such applications is Counterfactual Multi-Agent Policy Gradients (COMA), that employs difference rewards to provide each agent with an individual signal to learn from. This algorithm have however proved to perform poorly in practice: the reason for such poor performances has to be identified in the centralized critic used by COMA to estimate such values. Such a critic is difficult to learn because of compounding factors, and thus may provide inaccurate or wrong values to the agents. For this reason, in this work two novel algorithms, named Dr.Reinforce and Dr.ReinforceR, are proposed. These avoid the above difficulties by applying difference rewards on the system reward function, either by accessing it directly or by learning it with a centralized network. The results show the improvements over COMA, pointing out how learning a more accurate representation is a key factor towards a wider applicability of difference rewards to solve the multi-agent credit assignment problem. The original contributions presented in this thesis could deepen the understanding of multi-agent reinforcement learning, by providing evidences that carefully designed alternative representations are indeed useful in improving the learning of multiple agents and help in contrasting some fundamental problems that characterize this kind of systems. Moreover, novel techniques could build upon the solutions investigated in this work, and further improve performance or allow us to tackle increasingly complex problems.

The present dissertation takes late postwar Japan as its case study and investigates the ways in which ambivalence to/in motherhood’ emerges at the very site where maternal violence and, more specifically, maternal filicide disrupts social norms of acceptable maternal behaviour. In 1970s Japan the number of cases of mothers who killed their own children saw a dramatic increase to the point of reaching, within media representations in particular, the dimension of a social phenomenon. Within the framework of idealizations of maternal identity, formulated in terms of continuous love, self-sacrifice and domesticity, filicidal mothers came to be labelled as either "bad" (cruel, monstrous) or "mad" (mentally unstable, neurotic). The apparent proliferation of maternal child-killing and what was perceived as the unjust treatment meted out to these criminalized mothers became a major concern for a new women’s liberation movement emerging in Japan between the late 1960s and the early 1970s, known as ūman ribu (woman lib). Ribu contested the widespread characterization of mothers who kill as either devilish or mentally ill, and drew on the numerical increase of cases of maternal filicide as evidence of a symptomatic malfunctioning of the dominant gender ideology in modern Japanese society. Postwar Japan also witnessed a boom in women’s literature whose focus on the grotesque, on worlds of dreams and madness and on the morbid portrayal of female antisocial behaviours constituted fertile terrain for the proliferation of disquieting images of motherhood and maternal violence. This thesis focuses on the work by Japanese writer Takahashi Takako as a specific case study to address the discursive construction of filicidal mothers in women’s literature. This study acknowledges motherhood as a heated site of contested meanings and focuses on a close textual reading of media coverage, the rhetoric of ribu and women’s literature in order to explore the discursive constructions of mothers who kill which characterised early 1970s Japan. It sheds light on the problematic interactions between the different discourses under consideration and identifies the relationship between motherhood and violence as a hot-spot where clashing discourses produce a constant re-articulation of maternal and female identity.

This thesis is about automated reasoning in quantified modal and temporal logics, with an application to formal methods. Quantified modal and temporal logics are extensions of classical first-order logic in which the notion of truth is extended to take into account its necessity or equivalently, in the temporal setting, its persistence through time. Due to their high complexity, these logics are less widely known and studied than their propositional counterparts. Moreover, little so far is known about their mechanisability and usefulness for formal methods. The relevant contributions of this thesis are threefold: firstly, we devise a sound and complete set of sequent calculi for quantified modal logics; secondly, we extend the approach to the quantified temporal logic of linear, discrete time and develop a framework for doing automated reasoning via Proof Planning in it; thirdly, we show a set of experimental results obtained by applying the framework to the problem of Feature Interactions in telecommunication systems. These results indicate that (a) the problem can be concisely and effectively modeled in the aforementioned logic, (b) proof planning actually captures common structures in the related proofs, and (c) the approach is viable also from the point of view of efficiency.

This thesis explores the relationship between black gospel music and the African American freedom struggle of the post-WWII era. More specifically, it addresses the paradoxical suggestion that black gospel artists themselves were typically escapist, apathetic, and politically uninvolved—like the black church and black masses in general—despite the “classical” Southern movement music being largely gospel-based. This thesis argues that gospel was in fact a critical component of the civil rights movement. In ways open and veiled, black gospel music always spoke to the issue of freedom. Topics include: grassroots gospel communities; African American sacred song and coded resistance; black church culture and social action; freedom songs and local movements; socially conscious or activist gospel figures; gospel records with civil rights themes.

The growing visibility of the grassroots movement of men's support groups throughout this country and a resurgence of spiritual interest and faith within the American population come together in the emerging phenomena of men gathering together to worship. To better understand the motivations behind men's participation in such events, the Xavier Religious Index (XRI) was designed to identify factors forming underlying motivational constructs and was administered to two separate groups of men who had attended one or more such religious events. The test sample was comprised of 1190 men who attended a large, religious, men's conference. The cross validation sample was comprised of 301 men who had attended one or more men's retreat week-ends at a spiritual renewal center. In both samples, the factor identified as comprising the largest portion of the variance was that of Male Bonding. Other factors included: Self-awareness, Relationship with God, Isolation or Emptiness, Faith/Prayer Community, Father-Son Relationships, Coping Strategies, and Grief or Burdens. The data were also examined for differences across the life span. When the first and fourth quadrants of Sample 1 were grouped for age and compared, corresponding values existed. However, on examination of Factor 2, Self-awareness and Existential Concern, the directions in which these men focused were diametrically opposed, with older men endorsing a higher satisfaction with quality of life. The implications of this are discussed in light of developmental theory.