Fig. 3

An illustration of proposed causal graph learning framework using meta machine learning. We amortize the knowledge in the initialization graph for the causal structure learning problem \(\varvec{W}^{meta}\). For each task, we solve the structure learning problem with initialization from \(\varvec{W}^{meta}\). We also update this knowledge after solving each training task via meta-learning principles. For test task unseen, we adapt our knowledge to this task but do not update the knowledge