AIME2: toward a unified algebraic theory of explainability via approximate inverse operators | Synapse