Networks and graphs
To realize the potential of human centered AI, incorporating social competencies and social phenomena is a crucial step. In this sense, modeling of social phenomena is an important AI research topic. Drawing on ever-growing amounts of information encoded in electronic traces of social behavior, social interaction is often modeled in terms of networks of nodes and links. Real-world networks, including social networks and computer-, biological-, technological-, brain- and climate networks, are often complex: they have non-trivial technological features compared with simpler structures found in regular grids or random graphs.
Based on mathematical and physical modeling of networked systems and large scale experimentation, this collaboratory will make foundational contributions in the areas of:
Explainability: New schemes for explainable graph neural networks. Causal mobility models enabling counter-factual explainability and prediction of response to interventions. Application of social information to enhance the perceived relevance of AI explanations.
Novelty detection: Methods and algorithms for balancing accuracy and energy footprint of outlier detection at scale in complex network models.
Fair AI: Behavioral monitoring a the national scale with privacy guaranties and certified fairness for minorities.