Tue, 01 Dec 2026

15:00 - 16:00
C3

Mind the sign: what we miss when we ignore antagonistic interactions

Fernando Diaz-Diaz
(University of Madrid Carlos III)
Abstract

Many complex systems have both cooperative and antagonistic interactions. For example, countries form alliances and wage wars, financial assets move together or in opposite directions, genes activate or inhibit each other, and users of online platforms endorse or reject each other. We can represent these systems as signed networks, in which each edge has a positive or a negative sign. However, most tools of network science assume non-negative weights. The common practices of discarding negative edges, taking absolute values, or treating signs as ordinary weights can thus lead to wrong or incomplete conclusions. In this talk, we first review where signed networks appear and which problems arise when we apply standard methods to them. Then we discuss three types of analysis that account for signs explicitly: measuring structural balance at different scales, detecting mesoscale structure, and constructing network embeddings. Throughout, we illustrate how incorporating the signs into the analysis reveals structural properties that conventional approaches can overlook.

Tue, 24 Nov 2026

15:00 - 16:00
C3

Which links matter most? Sparsifying network dynamics with effective resistance

Prof. Cristopher Moore
(Santa Fe Institute)
Abstract
“Sparsification” is the act of reducing a network to a subset of its edges while approximately preserving its properties: either to reduce the computational cost of solving problems about it, or to identify which edges are the most important in some sense. Computer scientists have developed beautiful techniques for sparsifying a graph using physics-related ideas like the effective resistance. However, while these methods preserve the spectral properties of the Laplacian, it is not obvious to what extent they preserve the behavior of nonlinear dynamical systems. Using a mobility network from the United States as an example, I’ll show that they do very well for the SIR epidemic model, including the probability each node becomes infected and its distribution of arrival times, even when the sparse network includes less than 10% of the original edges. Choosing edges using purely topological methods, or by thresholding edge weights, does not perform nearly as well. I will end by discussing the possibility of using sparsification to “denoise” networks from bioinformatics, and present some preliminary results on the Kuramoto model of coupled oscillators.
 
This is joint work with Alexander Mercier (Harvard School of Public Health), Emmie Fitz-Gibbons (Brown), and Sam Scarpino (Northeastern).
Tue, 17 Nov 2026

15:00 - 16:00
C3

Embedding networks with the random walk first return time distribution

Vedanta Thapar
((Mathematical Institute University of Oxford))
Abstract

We propose the first return time distribution (FRTD) of a random walk as an interpretable and mathematically grounded node embedding. The FRTD assigns a probability mass function to each node, allowing us to define a distance between any pair of nodes using standard metrics for discrete distributions. We present several arguments to motivate the FRTD embedding. First, we show that FRTDs are strictly more informative than eigenvalue spectra, yet insufficient for complete graph identification, thus placing FRTD equivalence between cospectrality and isomorphism. Second, we argue that FRTD equivalence between nodes captures structural similarity. Third, we empirically demonstrate that the FRTD embedding outperforms manually designed graph metrics in network alignment tasks. Finally, we show that random networks that approximately match the FRTD of a desired target also preserve other salient features. Together these results demonstrate the FRTD as a simple and mathematically principled embedding for complex networks.

Tue, 10 Nov 2026

15:00 - 16:00
C3

Evaluation of discord and inference of real-world political opinions with the Voter Model

Dr. Antoine Vendeville
(MediaLab Sciences Po)
Abstract

Models of opinion dynamics describe how opinions are shaped in various environments. While these models are able to replicate general opinion distributions observed in real-world scenarios, their capacity to align with data at the user level remains mostly untested. In this talk, I will demonstrate that the MultiState Voter model with Zealots (MSVZ) can capture individual opinions in a fine-grained Twitter dataset collected during the 2017 French presidential elections. I will first lay out the theoretical background, with formulas to evaluate opinion fluctuation at the node level. I will also present exact equations to compute the probability of discord between nodes, that is, the probability that any two nodes hold different opinions. In particular, I will highlight how discord levels between two nodes hinge heavily on the statistical dependency between their states, which can be seen as a topological property of the underlying network. I will briefly illustrate how discord levels fluctuate in synthetic examples. I will then focus on the Twitter dataset, consisting of 19K accounts with pre-identified political leanings. Fitting the MSVZ on the retweet network with accounts of political parties as zealots, I will reveal a strong correspondence between individual opinion distributions in the equilibrium state and ground-truth political leanings. Additionally, I will demonstrate how discord probabilities can accurately identify pairs of like-minded users. These results depend heavily on taking into account the directionality and/or the weight of edges. Overall, this contribution advances our theoretical knowledge of the Voter Model, emphasizes its ability to identify individual characteristics in real-world datasets, and advocates for further empirical evaluations of opinion dynamics models at the user level.

Tue, 03 Nov 2026

15:00 - 16:00
C3

What is balance?

Joe Bacchus George
(University of Cambridge)
Abstract

Since the early 20th century, it was speculated that signed networks tend to display "structural balance". This is where certain configurations of positive and negative edges capture the property that a "friend of a friend is a friend" and the "friend of an enemy is an enemy". Originating in mathematical psychology for the purpose of studying triadic interactions, the notion of balance has since been generalised, and is believed to manifest outside of social systems.

Efforts have been made to quantify structural balance, which is rarely fully displayed in real world networks. In this talk, we will discuss current measures of balance and their fallacies, which tend to arise from computing what balance is not – as opposed to what it is. In turn, we propose a probabilistic model for structural balance, and suggest an inferential framework to robustly assess it's manifestation in real data.

Tue, 27 Oct 2026

15:00 - 16:00
C3

Permutation-Invariant Spectral Learning via Dyson Diffusion

Tassilo Schwarz
((Mathematical Institute University of Oxford))
Abstract

Diffusion models are central to generative modeling and have been adapted to graphs by diffusing adjacency matrix representations. The challenge of having up to $n!$ such representations for graphs with $n$ nodes is only partially mitigated by using permutation-equivariant learning architectures. Despite their computational efficiency, existing graph diffusion models struggle to distinguish certain graph families and their spectra, unless graph data are augmented with ad hoc features.
This shortcoming stems from enforcing the inductive bias within the learning architecture.
In this work, we leverage random matrix theory to analytically extract the spectral properties of the diffusion process, allowing us to push most of the inductive bias from the architecture into the dynamics. Building on this, we introduce the Dyson Diffusion Model, which employs Dyson's Brownian motion to capture the spectral dynamics of an Ornstein-Uhlenbeck process on the adjacency matrix.
Furthermore, conditioned on the spectral dynamics, we formulate a Lie group diffusion, appropriately modeling the remaining degrees of freedom. Strikingly, the resulting learning problem becomes permutation invariant at the Lie algebra level.
We demonstrate that the Dyson Diffusion Model learns graph spectra accurately and outperforms existing graph diffusion models.
This is joint work with Cai Dieball, Constantin Kogler, Renaud Lambiotte, Arnaud Doucet, Aljaz Godec, and George Deligiannidis.

Paper: Schwarz, T., Dieball, C., Kogler, C., Lambiotte, R., Doucet, A., Godec, A., Deligiannidis, G., 2026, Permutation-Invariant Spectral Learning via Dyson Diffusion, Accepted at NeurIPS 2026

Tue, 20 Oct 2026

15:00 - 16:00
C3

Topology shapes dynamics of higher-order networks

Prof. Ginestra Bianconi
(School of Mathematical Sciences Queen Mary University of London)
Abstract
Complex systems like the brain, climate, and next-generation artificial intelligence rely on interactions that extend beyond simple pairwise relationships. This talk reveals the critical role of the topology encoded by higher-order networks in shaping  their dynamics.
 
By combiniging algebraic topology with non-linear dynamics, this research highlights how topological signals, dynamical variables defined on nodes, edges, triangles, and other higher-order structures, drive phenomena such as Topological Synchronization. The surprising result that emerges from this research is that topological operators including the Topological Dirac operator, offer a common language for treating complexity, AI algorithms, and quantum physics.  
These findings not only advance the understanding of the underlying mechanisms in neuroscience and climate science but also pave the way for very innovative machine learning algorithms inspired by theoretical physics. 
 
Millán, A.P., Sun, H., Giambagli, L., Muolo, R., Carletti, T., Torres, J.J., Radicchi, F., Kurths, J. and Bianconi, G., 2025. Topology shapes dynamics of higher-order networks. Nature Physics, pp.1-9.
Tue, 13 Oct 2026

15:00 - 16:00
C3

Invasion graphs: the networks behind how ecological communities assemble

Dr. Violeta Calleja Solanas
(Department of Biology, Oxford University)
Abstract
Ask a complex systems’ researcher to draw an ecological network and you will likely get an interaction network: nodes are species, and links are who eats whom / who competes with whom / who pollinates whom, etc. But behind every ecological community hides a second network, the invasion graph. Each node is an entire community (a set of species able to persist together) and each directed link is an invasion: a newcomer species arrives, tries to establish, and sometimes extinguishes residents. Community assembly becomes a walk on this graph, and the graph’s structure records every history the community could have had. 
In this talk, I will show how to build these graphs from generalised Lotka–Volterra dynamics and present the relationships between the structure of the invasion graph and the structure of the interaction networks among species. 
Tue, 17 Nov 2026

12:30 - 13:00
C3

Modelling brain clearance across scales

Lucie Wintrová
(Faculty of Mathematics and Physics Charles University Prague)
Abstract

Brain activity produces waste proteins such as amyloid-β, whose accumulation is an early hallmark of Alzheimer’s disease. Amyloid-β  dynamics are governed by cell-scale processes, namely production at neurons, diffusion and aggregation in the extracellular space, and receptor-mediated transport across the blood–brain barrier. The resulting accumulation, however, plays out at the tissue scale.

In this talk, we introduce a multi-species reaction–diffusion model in a periodically perforated domain, which is coupled to nonlinear receptor  dynamics on capillary surfaces. Since the receptors do not diffuse, they  lack spatial regularity, which makes the analysis more delicate. We  prove well-posedness with uniform L∞ bounds and, using two-scale  convergence, derive an effective macroscopic model. Its diffusion tensor  comes from a cell problem, surface processes become volume sources, and  the receptors remain a cell-scale variable. Numerical experiments show  that, at fixed porosity, the shape and curvature of cells alone can change the effective transport by tens of percent.

Subscribe to C3