Dynamic Networks: Analysis and Modeling

DYNAMO is a research group at the University of California, Santa Barbara, led by Professor Ambuj Singh. Our research develops mathematical and computational foundations for learning, understanding, and reasoning with complex data. We bring together machine learning, graph algorithms, network science, geometry, and optimization to address fundamental questions in artificial intelligence and scientific discovery.

A central theme of our work is the relationship between representation and reasoning. How should complex data be represented so that meaningful structure can be discovered? What makes a representation robust, interpretable, and transferable across tasks? How does the geometry or symmetry of a representation influence what a model can learn and reason about? And how can we distinguish predictive success from genuine scientific understanding?

We investigate these questions through theoretical analysis, algorithm design, and empirical research across several interconnected directions.

Research Directions

Representation Learning, Geometry, and Symmetry. We study the mathematical structure of learned representations, including their geometry, invariances, and relationships across models and domains. Our goal is to understand how representations influence generalization, robustness, interpretability, and reasoning.

Graph Learning, Reasoning, and Explainability. Building on our foundations in graph algorithms and network science, we develop methods for learning from relational and structured data. Our work includes graph neural networks, graph representation learning, causal and counterfactual explanations, and the reasoning capabilities of large language models.

Human–AI Systems and Collective Intelligence. We investigate how people and AI agents interact, exchange information, influence one another, and make collective decisions. We study the dynamics of trust, cooperation, adversarial behavior, and collective judgment, combining computational models with empirical analysis.

AI for Scientific Discovery. We apply and extend foundational machine learning methods to problems in molecular science, drug discovery, and neuroscience. These applications motivate new representations and reasoning methods that capture physical, biological, and functional relationships, connecting data-driven learning with scientific knowledge.

Our Approach

Our research is guided by the belief that advances in AI require more than increasingly powerful predictive models. They require an understanding of the representations those models learn, the reasoning processes they support, and the conditions under which their conclusions remain reliable.

We combine mathematical foundations with computational methods and interdisciplinary collaborations. Our work moves between fundamental questions and scientific applications, using each to inform the other. Scientific problems expose limitations of existing learning methods, while advances in algorithms and theory create new opportunities for discovery.

People and Collaboration

DYNAMO brings together researchers with backgrounds in computer science, mathematics, and the sciences. We emphasize rigorous scientific thinking, interdisciplinary exploration, and the development of methods that contribute both to foundational AI research and to challenging problems in other disciplines.

Our work is reflected in our Research, Publications, and Team pages.