Past Talks

I will introduce the basic ideas to study fluctuations and rare events in physical sciences through the lens of large deviation theory. We will look at some examples and show how large deviations ideas are relevant to modern statistical physics. 

I will then share my experience as a PhD student and now as a postdoc on an independent fellowship, on which we can continue our discussion during the seminar.

The realm of graph neural networks (GNNs) has experienced remarkable and swift advancements in recent years. Emerging at the intersection of graph theory and deep learning, GNNs have permeated fields from biology to computational social sciences, revolutionizing complex data analysis. This presentation explores core concepts and advanced algorithms, discussing recent breakthroughs and open challenges in the field from both research and application perspective.

The convergence of complex systems studies and graph-based machine learning represents a rich soil for fostering innovation and collaboration. This presentation delves into their distinctive traits and shared elements, highlighting the synergy between them while delineating their respective strengths and weaknesses. Rather than just comparing them, in this seminar we aim to identify points of connection, providing strategies to facilitate their integration and advocating for a collaborative approach to scientific exploration.

A key challenge of nonlinear dynamics and network science is to understand how polyadic couplings influence collective behavior in complex systems. In this talk, I will show that the effects of higher-order interactions can be multifaceted by analyzing the dynamical patterns of identical Kuramoto oscillators on hypergraphs. In particular, higher-order interactions can have opposite effects on linear stability and basin stability, stabilizing twisted states by improving their linear stability, but also dramatically reducing their basin sizes, making them hard to find from random initial conditions.

In the second part of my talk, I will share some ideas involving scientists working in interdisciplinary research areas. I would like to discuss for example how to find your community as an interdisciplinary scientist, and share some of my thought about how to find a faculty job as an interdisciplinary scientist. In addition, I am happy to discuss any other questions.

Technological progress is growing by leaps and bounds, followed by new ethical dilemmas.

Is AI going to steal researchers jobs? Should researchers be rewarded in some way for training models?

In the first part of the talk, I will showcase the existing main tools that can facilitate research work. I will then discuss possible developments, what tools we might expect to see, and how the research landscape will change thanks to new technologies.

Finally, I will introduce ethical issues, which we will discuss in a virtual roundtable in the second part of the talk.

In the first part of the talk, I will present our generative model for creating synthetic random networks. We examined the topological features of real-world power grids, such as the average shortest path length, clustering coefficient, mean degree, and local maximum degree, in order to reproduce these properties in random systems. I would like to introduce the algorithm we have developed for this purpose, which takes the synthetic network's degree distribution as input and produces the corresponding synthetic network adjacency matrix as output.

During the second part of the discussion, I will tackle the challenge of communicating efficiently with a varied audience and effectively conveying technical research. Utilising visualisation tools for presenting research can be highly beneficial in catching the audience's attention and effectively transmitting information.

Natural selection governs the survival of the fittest, yet extensive cooperation can be observed worldwide, including within academia. In this seminar, I'll delve into our recent research on explosive cooperation in higher-order networks (https://arxiv.org/abs/2303.11475), addressing how cooperative behaviours emerge from competitive interactions. By extending pairwise games to higher-order networks, we identify a critical threshold of higher-order interactions and a critical initial cooperator mass triggering an explosive transition to cooperation, shedding light on the survival of cooperation.

In the second part, I'll offer a balanced perspective on volunteering in academia. On the one hand, volunteering helps to develop new skills and new connections, contributing to the whole academic community. On the other hand, balancing volunteer commitments with academic responsibilities requires effective time management and considering personal interests for engagement and satisfaction. We will hence discuss such benefits and considerations to uncover the multifaceted nature volunteering.

In the first part, I will present our latest results on the close link between response theory, reaction coordinates and critical phenomena for noisy systems with mean field interactions. Such systems are routinely used to model collective emergent behaviours in multiple areas of social and natural sciences as they exhibit, in the thermodynamic limit, continuous and discontinuous phase transitions.

Firstly, I will show how to define some reaction coordinates for the system starting from the coupling structure among the microscopic agents. Secondly, I will show that such reaction coordinates prove to be proper nonequilibrium thermodynamic variables as they carry information on correlation, memory properties and resilience properties of the system. In particular, the investigation of response properties of the reaction coordinates allows to identify, pinpoint and characterise phase transitions of the system as they manifest as singular values of the susceptibility associated to such thermodynamic variables. As an example I will apply our framework to the Kuramoto model and show how classical and new results can be obtained from a response theory perspective.

In the second part, I would like to discuss a few aspects of the PhD that I found somewhat challenging or, at least, a source of reflection and consideration. The macro-question I would like to discuss and hear other people’s opinions on regards the support structure during the PhD path (and beyond). A few micro-questions: What is the role of a research group? Is it useful/helpful? What is the role of a mentor/supervisor? How do you find a good balance between having enough independence and not feeling lost and whom should you ask if you are feeling lost? What is the best way to proceed in your career? Is the current PhD structure working?

I will talk a bit about my experience during the PhD, the challenges I faced and what has helped for my current research position.

In today's highly competitive academic environment we are faced with an ever-increasing daily deluge of new publications. Yet, only a small minority of such publications manage to attract significant attention. It is therefore of paramount importance to have a solid quantitative understanding of the determinants of academic impact, i.e., ultimately, of what factors may help a publication (and, more generally, its authors) to attract large numbers of citations. Despite well known controversies and criticisms, in most academic systems citations are in fact still very much used as the key indicator of academic impact. In this seminar I will review a number of results from my research that shed some light on academic impact. In particular, I will focus on long-lived academic careers, with the aim of understanding what factors or events that may occur during the early stages of an academic career may eventually translate into sustained impact in the long run. For instance, I will show how early career network effects, such as coauthoring papers with prominent scientists or collaborating with scientists from different disciplines, may have career-altering consequences. I will then show some results about historical trends in inequality (i.e., the concentration of citations across authors) and mobility (i.e., the possibility of climbing academic ranks) in scientific disciplines, concluding with some ideas and suggestions to improve the fairness of academic policymaking.

Online social platforms have become central in the political debate. In this context, the existence of echo chambers is a problem of primary relevance. These clusters of like-minded individuals tend to reinforce prior beliefs, elicit animosity towards others and aggravate the spread of misinformation. We study this phenomenon on a Twitter dataset related to the 2017 French presidential elections and propose a method to tackle it with content recommendations. We use a quadratic program to find optimal recommendations that maximise the diversity of content users are exposed to, while still accounting for their preferences. Our method relies on a theoretical model that can sufficiently describe how content flows through the platform. We show that the model provides good approximations of empirical measures and demonstrate the effectiveness of the optimisation algorithm at mitigating the echo chamber effect on this dataset, even with limited budget for recommendations.

In a second time, I would like to talk about the difficulty of juggling between microscopic and macroscopic research. As I tend to focus on small details instead of the larger picture, I spent too much time at the start of my PhD focusing on equations and technicalities. At the end of my first year I realised that I didn't have a clear idea of the scope and the end goal of my research. I've had quite a hard time taking a step back to look at my work on a more global level and identify what exactly it is that I am doing. I would be very interested to exchange on this matter and hear everyone else's thoughts.

Dark web marketplaces are forums where users can buy or sell illicit goods/services and transactions are typically made using cryptocurrencies. While there have been numerous coordinated shutdowns of individual markets by authorities, the ecosystem has been found to be immensely resilient. In addition, while transactions are open and visible by anyone on the blockchain, the sheer scale of the data makes monitoring beyond basic characteristics a huge effort. 

In the first part of this talk, I propose the use of temporal motif counting, as a way of monitoring both the system as a whole and the users within it. Focusing on the Alphabay and Hydra dark markets, I study all the motifs formed by three sequential transactions among two to three users, finding that they can tell us something more complex than can be captured by simply degree or transaction volume. Studying motifs local to the node, I show how users form salient clusters, which is a promising route for classification or anomaly detection tasks.

In the second part of the talk I discuss the challenges that the scale of the data posed in this work, as well as building algorithms on top of an evolving research tool. I will talk about how learning a little about data structures and algorithm complexity can go a long way, as well as knowing the basics of a code profiler, “rubber ducking” and finding a computer science friend who can speed up this learning curve with you.

To have kids with another person is one of the biggest choices that many people make in their life. Tendencies when making this partner choice with respect to socioeconomic backgrounds, age, and geography tells us a lot about the culture and structure of a society. The tendencies have been studied on a pair-wise level. However, there has been little any work done on their consequences seen from a societal level. In this talk, I will show some of my work on understanding the pair-wise tendencies of partner choices and how these affect the structure of society using network analysis.

I will also talk about a difficult task that I'm yet to master; how to find your footing and focus in a constant stream of new research opportunities. For me, this problem is closely tied to the knapsack problem of prioritizing my time and mental capacity.

The efficient transport of resources is a pivotal contributing factor in determining the structural evolution of real-world networks. Archetypes for self-organizing systems that ramify into networks in order to optimize energy expenditure rates are xylem conduits in leaves and river basins. However, these formations are not only restricted to the natural realm but can also be generated by anthropogenic processes. A prominent example is that of transportation networks such as railway and metro systems, that are designed to jointly optimize traffic overload and infrastructural cost. In this talk, I will talk discuss how these models can be studied with methods of statistical physics and optimal transport theory. These tools help unveil the macroscopic nature of mobility patterns and result in efficient algorithmic implementation for the simulation of navigability strategies in transportation.

After the talk, I will propose a discussion on "Dealing with the environmental impact of research".

With the hit of new pandemic threats, scientific frameworks are needed to understand the unfolding of the epidemic. The use of mobile apps that are able to trace contacts is of utmost importance in order to control new infected cases and contain further propagation. Here we present a theoretical approach that uses both percolation and message-passing techniques to quantify the role of automated contact tracing in mitigating an epidemic wave. Our approach captures the steady state of the SIR epidemic spreading with contact-tracing and test policies based on the mapping of the process to link percolation. Each individual is assigned a variable indicating the adoption or not of the app. Assuming perfect efficiency of the app, the model is based on the fact that only individuals with the app infected by individuals with the app are not able to spread the disease further if infected. Our study goes beyond previous attempts to study the effect of the app by fully capturing the non-linear effects of the dynamics and the phase diagram of the process. Moreover, we show that the adoption of the app by a large fraction of the population increases the value of epidemic threshold, and the best strategy in order to maximally delay the percolation transition is given by targeting the hubs. We use both percolation and message-passing techniques to study the role of contact tracing in mitigating an epidemic wave and we predict analytically the phase diagram of the model in random networks with given degree sequence. The analytical results are compared with extensive Monte Carlo simulations showing good agreement for  homogeneous, heterogeneous networks models and for real data. In conclusion, the proposed theoretical framework is able to assess the expected impact of contact-tracing apps in the course of an epidemic capturing the non-linear effect of the spreading dynamics. These results are important to quantify the level of adoption needed for contact-tracing apps to be effective in mitigating an epidemic.

In the second part of my talk, I will open a discussion about how to propose new research projects. I hope the member audience will share their experiences from distinctive perspectives (for instance the difference between a theoretical project and an applied project). I will also share some of my ideas. Overall, since clearly I don’t have much experience on proposing new projects myself, I hope this will be a discussion that can benefit us all.

In the last decade, temporal networks and static and temporal hypergraphs have enabled modelling connectivity and spreading processes in a wide array of real-world complex systems such as economic transactions, information spreading, brain activity and disease spreading. Here, we present the Reticula C++ library and Python package: A comprehensive suite of tools for working with real-world and synthetic static and temporal networks and hypergraphs. This includes various methods of creating synthetic networks and randomised null models based on real-world data, calculating reachability and simulating compartmental models on networks. The library is designed principally on an extensible, cache-friendly representation of networks, with an aim of easing multi-thread use in the high-performance computing environment.

In terms of challenges, I will talk more generally about the good and bad parts of writing and distributing software by scientists for scientists. What kind of skills would be useful? How can a PhD candidate reconcile scientific software development with the classic expectation of publishing papers and getting citations?

Hypergraphs and simplicial complexes both capture the higher-order interactions of complex systems, ranging from higher-order collaboration networks to brain networks. One open problem in the field is what should drive the choice of the adopted mathematical framework to describe higher-order networks starting from data of higher-order interactions. Unweighted simplicial complexes typically involve a loss of information of the data, though having the benefit to capture the higher-order topology of the data. In this work we show that weighted simplicial complexes allow to circumvent all the limitations of unweighted simplicial complexes to represent higher-order interactions. In particular, weighted simplicial complexes can represent higher-order networks without loss of information, allowing at the same time to capture the weighted topology of the data. The higher-order topology is probed by studying the spectral properties of suitably defined weighted Hodge Laplacians displaying a normalized spectrum. The higher-order spectrum of (weighted) normalized Hodge Laplacians is here studied combining cohomology theory with information theory. In the proposed framework, we quantify spectra of different dimension using higher-order spectral and compare the information content of higher-order entropies and spectral relative entropies. The proposed methodology is tested on real higher-order collaboration networks and on the weighted version of the simplicial complex model “Network Geometry with Flavor”.

In the second part of the talk, I will share some general difficulties that one can encounter when doing a PhD in Italy, especially regarding concerns about the perspectives at the end of the PhD, with a few words about salaries and rights of PhD students in Italy.

Cancer is the uncontrolled growth of abnormal cells in the body, and it develops from a single cell to bigger masses which take control of portions of healthy tissues. Following the Darwinian evolution theory, in order to successfully evolve, tumour cells are able to implement complex mechanisms for surviving and cheating the body’s normal control mechanism. The most efficient way to escape the human body defence is producing genetic mutations, by hiding oncogenes in small portions of DNA released outside the chromosomes, thereby relieving them of normal chromosomal constraints and enabling tumors to evolve at higher rates than normal cells. These DNA portions are called extra-chromosomal DNA structures (ecDNA), and they make unique contributions to oncogenic transcription and tumour progression through their promotion of oncogene amplification and treatment resistance. In light of the aforementioned considerations, investigating ecDNA evolutionary paths and mechanisms of reproduction could lead to a deep understanding of cancer progression and potential new therapeutic strategies. In this talk, I will then present a mathematical model that describes a multiple species ecDNA evolutionary process, and I will recall the most important results by giving an overview of the mathematical and numerical methods we used and then by showing and explaining our primary achievements in terms of distributional, dynamical and probabilistic aspects of ecDNA reproduction in different complex scenarios. Following also the NetPlace exciting talk schedule, I aim to open a discussion about the tougher difficulties I encountered and I am still encountering in my PhD journey and how this project has been realised during an exceptional time as the pandemic one.

During the covid epidemics, one of the crucial tasks people were trying to solve was to find the correct model that would best describe observed data. Learning such a model, would open the possibility to answer inference and prediction questions, like: what is the basic reproduction number or when it all started? Finally, one could use such a model for optimisation, in order to build future mitigation strategies or create efficient vaccination schemes. All of these problems are also relevant in the case of network models of spreading and current technology and reporting may allow us to learn such models from existing data. During my brief talk I will share some insight on statistical inference, learning and optimization for dynamics on networks. More specifically, I will focus on the realistic regime, where observed data is incomplete and subject to uncertainty. Such a setting makes the problem computationally demanding and requires approximations, when one is dealing with large networks. One such approximation technique is dynamic message passing, an inference method derived from belief propagation (cavity method), which is exact on tree graphs and very accurate for sparse graphs. I will also discuss difficulties that we encountered when working on the subject and some interesting outcomes, which resulted from them.

This is a bit of an experiment. At the beginning of the session, I will summarise some of the lessons I learned during my own time as a PhD student, and then later as a supervisor, about the process of doing a PhD. This is about the different stages of a PhD, the role of the supervisor and of the student in these stages, about frustration along the way, and why it is normal to struggle. This is based on these notes https://bit.ly/3HhSpHu. Hopefully, members of the audience will also contribute from their own experience. The meeting is meant to be interactive — I will try my best to avoid the impression that I have all the answers. Of course I don't, and of course everything I will say is highly biased by my own experiences, and won't apply to everyone. Overall, I hope that this will be a discussion meeting rather than me giving a talk.

From the spreading of infectious diseases to the diffusion of ideas, innovations, beliefs, and behaviors, contagion is one of the most fundamental dynamical processes in social systems. Many theories and models have been proposed to understand social contagion. However, existing models often have two major limitations - (1) Assumptions about contagion dynamics --- which have been challenged by recent empirical evidence, and (2) a disconnect between social contagion theory and fundamental cognitive mechanisms (like dissonance) governing human behavior.  

This talk hopes to address these limitations by proposing a model which unifies the theory of cognitive dissonance and social contagion. Such a model organically gives rise to a spectrum of contagion dynamics documented in the space of social contagion.

Moreover, I will share three difficulties I have encountered in this particular project:
(1)  Programming - shifting from python to julia.
(2) Thinking about cognitive science (a new field to me) while keeping the model simple enough.
(3) Empirical validation - this is ongoing work and I can point to some of the difficulties we are facing

Across various scenarios, individuals cooperate with others to contribute towards a shared goal and ensure self-preservation. In game theory, the act of cooperation is considered as an individual producing some form of benefit to be utilised by others, under the expectation others will return the favour. In several scenarios, individuals make use of their own information to aid with their decision about who to connect and cooperate with. However, the choice of cooperation can be taken advantage of by opportunistic defectors, which can lead to significant disruption. This paper investigates how the diversity of opinion can contribute to the structure and mechanics of a dynamical network model and to the resilience of cooperation, by utilising a computational model where individuals make use of both public and private information to implement their decision. Our results show that increasing diversity leads to more stable, less connected and less prosperous networks coupled to more frequent, but shallower information cascades. Our work generally shows that the outcome of the conflict between cooperators and cheaters strongly depends on the interplay between population structure, individual decision making and individual opinions.

I raise some questions and provide some advice for a number of methodological and justificatory concerns that arise during interdisciplinary research. There are many ways we might tell the "stories" of our research, so my central question is: Why ought my telling be the right one? In talking through a pragmatic answer, I draw upon ethnography, quantitative ethnography, analytic autoethnography, empirical ethics, and theories of coherence. I give a metaphor inspired by sticky notes on a desk, and I give a short introductory example based on my own analytic autoethnography work reflecting on my first year of hormone replacement therapy and reflecting on my role as a researcher.

Probabilistic latent variable models are popular models for learning patterns in networks, systems made of many individual elements interacting in pairs. Latent variables inferred from these types of data can provide useful insights, such as detection of communities or clusters of similar nodes. However, recent works have shown that these approaches fail to reproduce relevant structural network properties, for instance reciprocity, the tendency of two nodes to form mutual connections. We conjecture that this may be related to a common conditional independence assumption made in these approaches: edges are independent, conditional on the latent variables (e.g., community membership). In this talk, we present two variants of probabilistic generative models that relax this assumption while preserving the good properties of standard models. In particular, we present two distinct ways of relaxing conditional independence between edges: one that specifies conditional probabilities and relies on a pseudo-likelihood approximation and one that jointly models pairs of edges with exact 2-edge joint distributions. We show how these first attempts to break this assumption result in improved performance in terms of various inference tasks such as recovering communities and edge prediction. Remarkably, these models are now capable of generating synthetic networks that replicate the reciprocity values observed in real networks, thus overcoming the limitations of standard models and better serving the needs of practitioners interested in modeling this structural property.

This session will give an overview of my PhD research on residuality theory. The research involves viewing software architecture practices through the lens of the complexity sciences and proposes new techniques for making technical decisions in conditions of high uncertainty.  

In academia, and especially during a PhD, you might have struggled to have your work done and at the same time to have a work-life balance. Even you might have become so obsessed with self-improvement that you simply stagnated. At least that happened to me. That's why during my PhD I have been learning about how our concentration works and collected a bunch of resources and tips about dealing with procrastination. Maybe those are interesting for you! During this talk, I will revisit what we know about productivity and share some useful tools for you to stay less lost during this marathon. The content will inevitably reflect on my personal experience, and that's why then I encourage you to have a nice talk about your difficulties and own tricks.

Across income groups and countries, the public perception of economic inequality and many other macroeconomic variables such as inflation or unemployment rates is spectacularly wrong. These misperceptions have far-reaching consequences, as it is perceived inequality, not actual inequality informing redistributive preferences. The apparently universal phenomenon suggests the existence of a common mechanism behind public perceptions. We propose a network-based explanation of perceived inequality building on recent advances in random geometric graph theory. Our generating mechanism can simultaneously replicate the known stylised facts in the literature on biased perceptions of income inequality. Most importantly, we show analytically that homophilic graph formation implies a ‘middle class bias’, i.e., individuals perceiving themselves to be in the middle of the income distribution almost independently of their actual position. 

In a second step, we augment our model to account for perceptions of gender and racial wage gaps. We show that the combination of homophilic graph formation and estimation based on locally limited model is able to replicate both the underestimation of gender and racial wage gap that empirical studies find and also the well-documented fact that the underprivileged perceive the gender wage gap to be higher on average and with less bias. In contrast to these qualitative replication, we also demonstrate that the effect of homophilic graph formation is quantitatively too strong to replicate empirical estimates. We propose a simply remedy, where agents estimate using a composite signal based on local and global information. Our calibration suggests that the underprivileged place much more weight on the (correct) global signal than the privileged, in line with the well-established psychological finding that the adversely affected part of the population is more interested in global information about the issue. Our findings thus suggest that (educational) interventions about the global state of gender equality are much more likely to succeed than information treatments about inequality but that increasing diversity might aid in improving the accuracy of perceptions in both regards.

We study how the herd immunity threshold and the expected epidemic size depend on homophily with respect to vaccine adoption. We find that the presence of homophily considerably increases the critical vaccine coverage needed for herd immunity and that strong homophily can push the threshold entirely out of reach. The epidemic size monotonically increases as a function of homophily strength for a perfect vaccine, while it is maximized at a non-trivial level of homophily when the vaccine efficacy is limited. Our results highlight the importance of vaccination homophily in epidemic modeling.