Webinar: Learning Time-Varying Graphs from Online Data (video)
Graphs provide a powerful way to represent relationships between interacting entities, but in many real-world systems these relationships are not fixed. They can evolve as new observations become available, making it important to develop methods that can learn and track graph structures over time.
In this webinar, the presenter will revisit the problem of learning time-varying graphs from online data. Starting from the intuition behind graph learning and the challenges introduced by evolving network structures, the presentation will introduce a general framework for adapting graph estimates as new data arrives. The approach is designed to be model-independent, making it applicable to several common graph-based models and allowing different assumptions about the relationship between the data and the underlying graph to be incorporated into the learning process.
The presentation will focus on the main ideas and intuition behind the proposed methodology, rather than on technical details alone. He will also discuss the experimental results and illustrate how online graph learning can be used to capture changing relationships in networked data. The goal is to provide an accessible overview of the framework and its broader relevance to signal processing and machine learning on graphs.
