Francesca Pistilli "Structured Representations for Long-Horizon Procedural Reasoning"

This talk can be viewed via zoom. (Note: this talk will be recorded)

Abstract: Understanding how environments evolve over time is a fundamental challenge for intelligent systems, particularly during activities involving long sequences of interdependent actions and state changes. While recent generative models have made remarkable progress in predicting future visual observations, modelling complex procedures requires understanding not only what happens, but also how entities and their relationships change over time.

Structured representations, such as dynamic scene graphs, can offer a way to explicitly capture how environments evolve. Combining hierarchical representations of high-level task progression with fine-grained models of local state transitions provides a promising direction for reasoning over long-horizon procedural activities.

Surgical procedures offer a particularly challenging setting for investigating these approaches, as anatomical structures are non-rigid and undergo complex transformations. More broadly, structured models of procedural evolution could open new directions for understanding and assessing procedural execution, moving beyond individual action recognition towards reasoning about the progression and consistency of entire activities.

Bio: Francesca Pistilli is a researcher at Politecnico di Torino, Italy, working at the intersection of computer vision, graph representation learning, and robotics. Her research focuses on structured representations of scenes and human activities, investigating how graph-based models can help intelligent systems understand interactions, reason about changes in dynamic environments, and support robotic tasks. Her work spans 3D scene understanding, egocentric video analysis, and dynamic scene graphs.

She received her Ph.D. in 2023 from Politecnico di Torino, where she worked on graph neural networks and geometric representations for 3D point cloud understanding. She also holds an M.Sc. in Electronic Engineering from Politecnico di Torino and an M.Sc. in Electrical and Computer Engineering from the University of Illinois Chicago.