CSXAI COMPLEXITY SCIENCE FOR UNDERSTANDING AI: FROM THE DYNAMICS OF COMPLEX NETWORKS TO COLLECTIVE EFFECTS OF INTERACTING NEURAL NETWORKS
The CSxAI project engages researchers at IFISC in a (non-oriented research project) on the interdisciplinary challenge of applying tools and concepts of complexity science (dynamical systems, statistical physics, and network science) to better understand and model the inner workings of artificial neural networks and machine learning solutions. There is indeed an urgent need in Machine Learning (ML) for new ideas that can provide understanding and mechanistic interpretability of how and why machine learning models -including deep learning ones- work. The starting hypothesis of this project CSxAI is that, precisely,
tools and concepts from Complexity Science can be of great help for this endeavor, and researching the interface between Complexity science and
Machine Learning represents an opportunity for both fields. This project has two main objectives. Objective 1 (Towards a Dynamical Systems Toolkit for
Temporal Networks) aims at characterizing Temporal Networks i.e. networks whose architecture changes and evolves over time- using the paradigm and
tools of time series analysis and dynamical systems theory. This new focus on temporal network theory predicates on interpreting temporal networks as network trajectories, i.e. orbits from a latent graph dynamical system.
The second objective of the project is to describe Network trajectories in machine learning training (WP2) and to research collective phenomena in swarms of interacting brains (WP3).
Investigadors
Juan Fernández Gracia
Annalisa Caligiuri
Pedro Jiménez