Graphs for artificial neural networks and brain connectivity exploration

The main objective of this thesis is to explore brain and artificial neural network connectivity from a graph-based perspective. While structural and functional connectivity analysis has been extensively studied in the context of the human brain, there is a lack of a similar analysis framework in artificial systems. To address this gap, this research focuses on two main axes. In the first axis, the main objective is to determine a healthy signature characterization of the human brain's resting state functional connectivity. A novel framework is proposed to achieve this objective, integrating traditional graph statistics and network reduction tools to determine healthy connectivity patterns. Hence, we build a graph pair-wise comparison and a classifier to identify pathological states and rank associated perturbed brain regions. Additionally, the generalization and robustness of the proposed framework are investigated across multiple datasets and variations in data quality. The second research axis explores the benefits of brain-inspired connectivity exploration of artificial neural networks (ANNs) in the future perspective of more robust artificial systems development. A major robustness issue in ANN models is represented by catastrophic forgetting when the network dramatically forgets previously learned tasks when adapting to new ones. 
Our work demonstrates that graph modeling offers a simple and elegant framework for investigating ANNs, comparing different learning strategies, and detecting deleterious behaviors such as catastrophic forgetting. Moreover, we explore the potential of leveraging graph-based insights to effectively mitigate catastrophic forgetting, laying the foundations for future research and explorations in this area.

Président:

Nadia Brauner ()

Directeurs:

• Sophie Achard
• Michel Dojat

Raporteurs:

• Bertrand Thirion
• Nicolas Farrugia

Examinateurs:

• Ana Marques
• Giulia Preti