When? 1 June, 09:00 - 12:30
Where? Kendall Square Meeting Room (Lobby Level), Hyatt Regency Boston / Cambridge, USA
We are proud to present a high-quality program with speakers from different communities. Speakers who want to publish their talks after the satellite can send their slides to the organizers via E-Mail: Maja Lindström, Chester Tan, or Chen Zhang.
| JUNE 1st | Presenter |
|---|---|
| 09:00 - 09:10 | Opening Remarks |
| 09:10 - 10:00 |
Bruno Ribeiro
Purdue University, USA Keynote — Title: Towards Higher-Order Predictive Joins: A Missing Primitive for Agentic Relational AI Agentic AI systems operating over real-world relational data must answer queries whose results do not exist in any database. Standard database operations retrieve rows that are present — executing joins over existing foreign-key relationships. But the most consequential agentic queries are predictive joins: given a relational schema with tables for suppliers, components, plants, and products — or for students, projects, grants, and deadlines — which k-tuples of entities across these tables will yield a missing or anomalous high-order joint? Such queries admit no expression in relational algebra, and no solution by decomposing them into independent pairwise predictions. I will first establish a fundamental impossibility result: node embeddings capture only marginals of the joint positional distribution, making recovery of the joint distributions required for order-r predictive joins impossible from traditional node embeddings alone. To overcome this barrier, I introduce Holographic Representations — a variable-size node embedding framework that enables task-agnostic pre-training of expressive predictive models across any join order. Building on these foundations, I introduce new methods for in-context learning over relational graphs, enabling a pre-trained agentic model to execute novel predictive joins on unseen datasets at inference time — generalizing across graph structure, relation types, attribute domains, and join order from a small set of labeled examples. Pushing further, the same pre-trained models support zero-shot transfer to relational schemas with entirely distinct attribute spaces, decoupling learned structural dependencies from the specific feature domains found during training. Together, these results establish higher-order predictive joins as the natural primitive for agentic relational AI — and higher-order network theory as the essential foundation for the next generation of AI relational agents. |
| 10:00 - 10:30 |
Vincenzo Perri
ISI Foundation, Turin, Italy Title: Analyzing Sequential Group Interactions in Temporal Networks We investigate the dynamics of interactions between sets of nodes in sequential data, focusing on patterns that emerge when groups follow, overlap with, or evolve into one another over time. Such structures arise naturally in communication, collaboration, mobility, biological, and information systems, where events often involve multiple entities simultaneously. We consider methods to represent these sequential set interactions, compare observed patterns against suitable baselines, and assess their implications for downstream dynamics and prediction. |
| 10:30 - 11:00 | Coffee Break |
| 11:00 - 11:30 |
Timothy LaRock
Princeton University, USA Title: Comparing Observed and Simulated Human Mobility Paths with Higher-order Networks The detailed study of individual human mobility requires large-scale high-resolution datasets, but collecting such datasets in a way that is both statistically powerful and privacy preserving is a challenging and expensive task. In response, researchers have built tools to generate complex synthetic populations of agents that can be used to simulate synthetic individual mobility data, potentially obviating the difficulties of data collection. While these simulation-based approaches offer a promising avenue for expanding individual mobility research, it is difficult to assess whether such tools are effective at generating realistic mobility traces. In this work, we develop a framework for comparing observed and simulated mobility data using a higher-order network framework that focuses on analyzing patterns of movement in the paths individuals take through the underlying infrastructure network. We apply our framework to a case study comparing the NetMob 2025 Data Challenge Dataset, which includes individual mobility data for thousands of residents of the Île-de-France region, with a sophisticated open-source synthetic population and mobility simulation model of the same region. We show that while simulated mobility data is indeed promising as a surrogate for observed mobility, there are some key limitations to the simulation paradigm from a path-based perspective, which we discuss along with potential future remediations and open challenges for higher-order mobility network analysis. |
| 11:30 - 12:00 |
Vaiva Vasiliauskaite
ETH Zürich, Switzerland Title: Directed Cycles as Higher-Order Units of Information Processing in Complex Networks Directed cycles are fundamental motifs in many natural, social and artificial networks, yet their distinct computational roles remain under-explored, particularly in the context of higher-order structure and function. In this work, we investigate how two types of directed cycles — feedforward and feedback — can act as higher-order structures to facilitate the flow and integration of information in sparse random networks, and how these roles depend on the local and global environment of the cycles. Using information-theoretic measures, we show that network size, sparsity and relative directionality critically impact the information-processing capacities of directed cycles. In a network with no-preferred global direction, a feedforward cycle enables greater information flow, and a feedback cycle allows for increased information integration. The relative direction of a feedforward cycle, as well as the structural incoherence it induces, determines its capacity to generate higher-order behaviour. Finally, we demonstrate that introducing feedback loops into otherwise feedforward architectures increases the diversity of network activity patterns. These findings suggest that directed cycles serve as computational motifs with local information processing capabilities that depend on the structure they are embedded in. Using directed cycles, we highlight the interdependence between higher-order structures and the higher-order order behaviour they can induce in the network dynamics. |
| 12:00 - 12:20 | Panel Discussion https://app.sli.do/event/oZSKixAjRo8Hj6ogLnoJrL |
| 12:20 - 12:30 | Closing Remarks |