Social media was initially expected to generate widespread improvements in individual well-being and to have positive externalities in the political domain. Such optimistic expectations soon gave way to a host of concerns about possible negative effects of social media. At the societal level, many started worrying that social media might exacerbate political polarization and propagate misinformation. At the level of individual users, many became concerned that social media might have detrimental effects on mental health, especially among the youth. This research proposal has two major goals: first, to fill critical gaps in our knowledge of the causal effects of social media on users and society; second, to evaluate the effectiveness of interventions aimed at mitigating the downsides of social media. On the political front, I will deploy a novel empirical strategy to investigate the impact of social media on political engagement and conduct a large-scale field experiment to evaluate the effectiveness of an intervention aimed at increasing the quality and reducing the partisanship of the news consumed on social media. On the users’ front, I will conduct two large-scale field experiments: one aimed at evaluating the impact of an intervention encouraging parents to delay the acquisition of a smartphone for their children on outcomes related to child development; the other aimed at evaluating whether tweaks to the social media environment can mitigate the negative effects of social media on mental health. The overarching goal of this research project is to generate new scientific knowledge on the use and impact of social media, to highlight potential areas of policy intervention, and to evaluate ecologically valid and scalable remedies to the downsides of social media. The unique blend of cutting-edge causal inference techniques from observational data and large-scale field experiments will generate insights that can benefit researchers, users, and policymakers alike.

I will analyse how agents’ traits affect behaviour and strategic reasoning in decision problems and games. Traits encompass stable cognitive and psychological features of agents, such as tastes, skills, cognitive abilities, attitudes toward uncertainty and misspecification, concerns for others, and propensities to be affected by emotions. To build an adequate formal framework, I will try to comply with a separation principle: the description of the rules of interaction should be separate and independent from the description of players’ traits. This will enable a fruitful and conceptually rigorous analysis of personal traits in games, including players’ abilities and cognition, analogous to what has already been done—for example— concerning attitudes toward risk. Also, to model interactive strategic thinking (a.k.a. epistemic game theory), I will use a flexible approach that, unlike the standard one, does not assume common knowledge of cognitive rationality (e.g., of coherence). I will start with individual decision making and planning and then embed the analysis in interactive situations, with a special attention to sequential decision making and the role of time. Building on and improving upon my previous work on the foundations of game theory and psychological games, most of the analysis of the impact of traits on behaviour will be either focused on steady-state long run outcomes, using variations of the self-confirming concept, or outcomes consistent with subjective rationality and strategic reasoning, characterized by variations of rationalizability. Examples of the research questions to be addressed are the following: (1) How do cognitive feature affect strategic interaction? (2) Can institutions and agreements be robust to assumptions about interactive knowledge and beliefs about traits when players reason by forward induction? (3) How can we model context-dependent motivations? (4) What is the impact on long-run of concerns for misspecification?

In a recovery problem, we are interested in recovering structure from data that contains a mix of combinatorial structure and random noise. In a robust recovery problem, the data may contain adversarial perturbations as well. A series of recent results in theoretical computer science has led to algorithms based on the convex optimization technique of Semidefinite Programming for several recovery problems motivated by unsupervised machine learning. Can those algorithms be made robust? Sparsifiers are compressed representations of graphs that speed up certain algorithms. The recent proof of the Kadison-Singer conjecture by Marcus, Spielman and Srivastava (MSS) shows that certain kinds of sparsifiers exist, but the proof does not provide an explicit construction. Dynamics and population protocols are simple models of distributed computing that were introduced to study sensor networks and other lightweight distributed systems, and have also been used to model naturally occurring networks. What can and cannot be computed in such models is largely open. We propose an ambitious unifying approach to go beyond the state of the art in these three domains, and provide: robust recovery algorithms for the problems mentioned above; a new connection between sparsifiers and the Szemeredi Regularity Lemma and explicit constructions of the sparsifiers resulting from the MSS work; and an understanding of the ability of simple distributed algorithms to solve community detection problems and to deal with noise and faults. The unification is provided by a common underpinning of spectral methods, random matrix theory, and convex optimization. Such tools are used in technically similar but conceptually very different ways in the three domains. By pursuing these goals together, we will make it more likely that an idea that is natural and simple in one context will translate to an idea that is deep and unexpected in another, increasing the chances of a breakthrough.

Recent years have seen a rapid increase in available information. This has created an urgent need for fast statistical and machine learning methods that can scale up to big data sets. Standard approaches, including the now routinely used Bayesian methods, are becoming computationally infeasible, especially in complex models with many parameters and large data sizes. A variety of algorithms have been proposed to speed up these procedures, but these are typically black box methods with very limited theoretical support. In fact empirical evidence shows the potentially bad performance of such methods. This is especially concerning in real-world applications, e.g. in medicine. In this project I shall open up the black box and provide a theory for scalable Bayesian methods combining recent, state-of-the-art techniques from Bayesian nonparametrics, empirical process theory, and machine learning. I focus on two very important classes of scalable techniques: variational and distributed Bayes. I shall establish guarantees, but also limitations, of these procedures for estimating the parameter of interest, and for quantifying the corresponding uncertainty, within a framework that will also convince outside of the Bayesian paradigm. As a result, scalable Bayesian techniques will have more accurate performance, and also better acceptance by a wider community of scientists and practitioners. The proposed research, although motivated by real world problems, is of a mathematical nature. In the analysis I consider mathematical models, which are routinely used in various fields (e.g. high-dimensional linear and logistic regressions are the work horses in econometrics or genetics). My theoretical results will provide principled new insights that can be used, for instance in multiple specific applications I am involved in, including developing novel statistical methods for understanding fundamental questions in cosmology and the early detection of dementia using multiple data sources.