For many scientific domains, it is important to model the relationship between structure and function of systems with complex interactions among its parts. Graphs are mathematical objects that capture connections between parts of a system, providing a natural language for this. Examples can be found across all levels of neuroscience: chemical structures, network connectivity, functional interactions between brain regions, cognitive models of relational reasoning, and collective behaviors. Machine learning innovation on these models has been rapid in recent years, particularly with the advent of graph neural network methods, and their application to neuro-adjacent domains (e.g. chemistry, molecular biology, dynamical systems) has illustrated their potential. We believe these models can contribute significantly to progress in multiple neuroscience specialties, but overwhelmingly members of the Cosyne community seem to not know about them. This workshop seeks to correct this by highlighting recent applications of graph-based methods in different areas of neuroscience, and by bringing together experimentalists with graph-structured data and theorists with expertise in graph methods. Given the breadth of applications of graph-based approaches, we believe our workshop offers significant potential for catalyzing new research directions for members of the Cosyne community.
Organizers: Brian DePasquale (BU); Kim Stachenfeld (Deepmind & Columbia), Sam Lewallen
Nina Miolane
UCSB
Beyond graph neural networks: A survey of message passing topological neural networks for neuroscience
Kristin Branson
Janelia Research Campus
Training and interpreting GenAI models of animal behavior
Quan Do
Boston University
A graph neural network framework to model human reasoning
Dominique Beaini
MILA, University of Montreal, Valence Discovery
How to learn a molecule
Ashok Litwin-Kumar
Columbia University
Graph embeddings for identifying symmetries in connectomes
Gal Mishne
UCSD
Learning the dynamics of functional connectivity
Wesley Qian
Osmo
Osmo: Digitizing smell with a principal odor map
Stephan Saalfeld
Janelia Research Campus
Graph neural networks uncover structure and function underlying activity of neural assemblies
Schedule
Morning session
9:30 - 9:40: Welcome and introduction
9:40 - 10:05: Ashok Litwin-Kumar
10:10 - 10:35: Gal Mishne
10:40 - 11:05: Quan Do
11:10 - 11:30: Coffee break
11:30 - 11:55: Wesley Qian
12:00 - 12:25: Stephan Saalfeld
Afternoon session
3:30 - 4:15: Mini-tutorial on GNNs in PyTorch Geometric
4:15 - 4:40: Nina Miolane (virtual)
4:40 - 5:00: Coffee break
5:00 - 5:25: Dominique Beaini
5:30 - 5:55: Kristin Branson
6:00 - 6:30: Discussion