The Women in Network Science (WiNS) seminar is an interdisciplinary seminar with the aim to promote and showcase research by women and nonbinary researchers in network science. 

The seminar is open to everyone. Please join the mailing list to receive announcements and zoom links for upcoming seminar talks.

Elena Candellone, Haily Merritt, Gülşah Akçakır, Katherine Hamilton, and Ketika Garg convene this seminar series. Please get in touch if you are interested in presenting in the seminar or if you would like to nominate speakers.

For all scheduled talks, relevant preprints are available on our ZeroDivZero repository. Recordings of past talks can also be found in the ZeroDivZero repository and on Youtube.

Scheduled talks
Mondays - 11 am Eastern Time - on Zoom 

Bio: Laura W. Dozal is a PhD Candidate at the University of Arizona’s College of Information (iSchool). She is a Computational Social Scientist with 6+ years of specialization in information behavior and visual analysis through Computer Vision, Natural Language Processing, Network Science, and knowledge discovery. Her research methods focus on exploring online social movement messaging sentiment through domain and open science expertise. Recently she's been interested in identifying scenarios to produce synthetic multimodal data using ML/AI methods for larger downstream tasks.

Abstract: This research project answers the question; how can we identify a narrative structure from a collection of topical images using computational methods. The analysis is completed in a three-step AI/ML pipeline to build a narrative structure representative of the local semantics of image groups and an overall global summarization of the narrative structure as a whole. This first step pulls feature embeddings from a collection of 16,657 Instagram images pertaining to the anti-feminicide movement in Mexico; and clusters them to identify common vector embedding groups within the data. These images have multilingual text and contain various types of visual content. Human-in-the-Loop (HITL) evaluation analysis is applied to provide a qualitative understanding of the images within each cluster group and gain insight on the classification aspects of the images for downstream tasks. The next step takes these images and runs them through foundational vision transformer models to generate two labels identified in the HITL analysis, and a description of each image. The outputs are compared and analyzed using multi-label classification metrics for machine learning models along with HITL discourse analysis of the categorization process. Step three uses the top labels created from the model with the best overall quantitative and HITL discourse evaluation metrics to build a network graph representing the various stages of semantics found within the image and its attributes: 1) labels 2) descriptions and original post comments. The graph is analyzed to identify structural patterns of the network using traditional centrality measures and a modularity algorithm (Leiden). Variational graph autoencoders using a graph attention network were run to integrate sentiment into the network’s structure and understand the narrative as a whole while taking into account its local features.

Bio: Dr Assemgul Kozhabek is a Postdoctoral Research Associate at Heriot-Watt University and an Associate Fellow of the Higher Education Academy. She previously worked as a Data Scientist and Research Associate at Ulster University, where she applied machine learning and data analytics to industry-focused projects in smart manufacturing and robotics. Her research centres on network science, data analytics, and computational modelling of complex systems, with applications in urban infrastructure, scientific complexity, and research reproducibility. She holds a PhD in Computer Science from Bournemouth University and an MSc in Engineering Business Management from the University of Warwick. Outside academia, Assemgul enjoys exploring local destinations through slow travel and is committed to mindful, sustainable consumption in her everyday life.

Abstract: Urban road systems can be understood as complex networks whose structure strongly shapes their resilience and traffic dynamics. We analyse urban road networks using a multi-scale network science framework, examining global, meso-scale, and nodal properties across densely populated cities. At the global scale, network metrics capture efficiency and robustness, while meso-scale analysis reveals strong community structure and weak core–periphery organisation, indicating predominantly polycentric urban forms. Nodal-level centrality measures exhibit distinct yet consistent distributional patterns across cities. To assess robustness, we evaluate the impact of targeted and random node removal strategies on network connectivity and efficiency. Centrality based disruptions are shown to be significantly more effective than random failures, with shortest-path-based measures most damaging to global connectivity and local-connectivity measures primarily affecting local efficiency. Counter-intuitively, some perturbations increase local efficiency, highlighting non-trivial resilience effects. Finally, we model traffic congestion as a network-constrained spreading process using a topology-aware SIR framework. Applied to real-world traffic data, the model outperforms classical mean-field approaches, offering improved insight into congestion propagation.

Bio: Yasaman Asgari is a PhD student in the Department of Mathematical Modeling and Machine Learning and the Digital Society Initiative at the University of Zurich, Switzerland. Her research focuses on network science and its applications to the science of science and social science.

Abstract: Who studies whom in global science and how does this attention shift in response to major world events? In this talk, I address this question by analyzing over 36 million scholarly journal articles published between 2000 and 2022 and extracting country mentions to construct a novel network representation, which we refer to as a scholarly attention network. Focusing on the Arab Spring, we examine how sociopolitical events reshape the global distribution of scholarly attention across 10 countries in the Middle East and North Africa. We show that attention toward affected countries increased significantly, and besides Western countries, Saudi Arabia and Malaysia contributed the most to this increase. Moreover, these shifts can be partially explained by changes in research funding and scholarly migration, linking attention dynamics to broader structural forces in science.

Bio: TBD

Abstract: Complex problem-solving requires navigating a fundamental tension between exploring new possibilities and exploiting existing knowledge to find optimal solutions. Networked systems often solve this trade-off remarkably well, and prior work has linked structural features such as small-world topology to efficient collective performance. Yet the underlying causal information dynamics remain unclear. In this talk, I will present results from an agent-based model of combinatorial search, the Potions Task, and introduce an information-theoretic framework that quantifies exploration–exploitation in networks through measures of redundancy and synergy in agents’ discoveries. Through this framework, we quantify how this dynamic unfolds over time and structures, and show that informational synergy drives performance across network structures and can override structural advantages . These results suggest a potentially general principle governing decentralized collective problem-solving, and highlight how network science can move beyond topology to uncover the informational mechanisms that drive collective intelligence.

Bio: Linnéa Gyllingberg is an applied mathematician working across disciplinary boundaries to develop mathematical and computational models of complex systems. She received her PhD in Applied Mathematics from Uppsala University, Sweden, in 2024, and began her postdoctoral research at UCLA as a Fulbright Scholar. She is currently a Wallenberg Postdoctoral Fellow in the Department of Mathematics at UCLA. Her research focuses on biological intelligence, adaptive behavior, and the theory and practice of scientific modelling. She develops mathematical models combining nonlinear dynamics, adaptive networks, and spatial modelling to understand how intelligence and adaptive behaviour emerge in living systems, and investigates the broader role of mathematical modelling in biological theory building.

Abstract: Across living systems, oscillations support coordination, information flow, and decision making, from neural rhythms in brains to calcium signaling in single cells. The unicellular slime mould Physarum polycephalum is a striking example: despite lacking a nervous system, it can navigate through mazes, form efficient transport networks, and balance exploration with exploitation. These behaviors are driven by oscillatory contractions that propagate through a dynamic transport network, which reorganizes as the organism adapts to its environment. However,most mathematical models of Physarum focus either on local oscillatory dynamics or on large-scale network adaptation, treating structure and dynamics separately. In this talk, I explore how oscillatory dynamics and adaptive network structure can be coupled within a single framework. By linking local biochemical oscillations to global network reorganization, the model generates symmetry breaking, directed transport, and large-scale redistribution of mass. These results suggest that decision-like behavior can emerge from nonlinear feedback between oscillatory dynamics and evolving network structure. More broadly, this work highlights oscillatory adaptive networks as a mechanism for distributed biological information processing without centralized control.

Bio: Carrisa Cocuzza earned her PhD in Neuroscience at Rutgers University in 2022 under the advisement of Dr. Michael W. Cole. Her dissertation examined the extent that brain-network-based mechanisms can explain local and distributed processes, including visual category selectivity and cognitive control. She is now working as a Postdoctoral Fellow in collaboration with Dr. Avram J. Holmes at the Center for Advanced Human Brain Imaging Research at Rutgers University, along with collaborative, multi-institute training at Princeton University’s Latent Cause Inference Conte Center. Carrisa’s work incorporates computational approaches from machine learning and network control theory, along with multivariate sources of biological and behavioral data, to build mechanistically-informed models of transdiagnostic neurocognitive deficits.

Abstract: Human cognition emerges through flexible and dynamic large-scale brain network interactions – an aspect of temporal coordination that may be transdiagnostically impaired across psychiatric illnesses. While substantial progress has been made studying neurobiological factors contributing to symptom expression, research has largely relied on ‘static’ patterns of brain connectivity. Despite the importance of understanding links between temporal characterizations of brain function and behavior, the mechanistic relevance of dynamic network organization to disease remains to be established. Here, in 203 participants (n=129: affective/psychotic diagnoses, n=74: no diagnosis), whole-brain functional connectomes from resting- and task-state fMRI were decomposed to identify dynamical constraints on across-state network interactions. Dynamical constraints were less variable across task contexts in patients, suggesting that de-differentiated (i.e., flattened) network dynamics are linked with failures to meet cognitive demands. We applied dimensionality reduction and hierarchical clustering to over 100 self-report and clinical measures spanning psychopathology to uncover clusters of phenotypic dimensions. After converting each participant’s pattern of cluster expression into a phenotypic fingerprint, across-state network dynamics were able to classify dimensional phenotypes more robustly than traditional case-control status. Further, the extent that dynamics were de-differentiated was positively and negatively linked with externalizing features in patients and healthy controls, respectively. Graph theory revealed that a role of dynamical constraints on brain network reconfiguration may be enabling efficient information flow over shifting connectivity patterns. We found novel evidence that brain network dynamics are linked to prominent dimensions of the phenotypic hierarchy exhibited across healthy controls and transdiagnostic psychiatric patients.