Identifying the Group to Intervene on to Maximise Effect Under Cross-Group Interference

A research team from University of South Australia and University of Technology Sydney has published a breakthrough paper addressing a fundamental challenge in networked systems: cross-group interference. When interventions (like public health campaigns or digital marketing) are applied to one group, their effects cascade through complex network connections to impact other groups. The researchers formalized this as the "cross-group causal influence estimation" problem and introduced Co2G (core-to-group causal effect), a mathematically rigorous way to measure these cascading impacts.

To solve this computationally intensive problem, they developed CauMax, an uncertainty-aware causal effect maximization framework. CauMax employs two scalable algorithms: CauMax-G uses iterative greedy search with Monte Carlo dropout for confidence bounds, while CauMax-D applies differentiable gradient-based optimization via Gumbel-Softmax relaxation. Both approaches leverage graph neural networks to model interference patterns across groups. In extensive experiments on real-world social networks, CauMax demonstrated remarkable performance, achieving an order-of-magnitude (10x) reduction in regret compared to traditional structural heuristics and diffusion-based baselines. The research also found that moderate uncertainty penalization consistently improves subset selection quality.