Comment by youknowme123

This is nice, it reminds me of earlier work where DAG-based optimization was used to discover effective multi-agent architectures under supervised learning.