$\partial$CBDs: Differentiable Causal Block Diagrams

Researchers have introduced a new modeling framework called Differentiable Causal Block Diagrams. It combines three previously separate goals: modular system design, gradient-based learning from data, and formal verification for safety. This allows engineers to create complex cyber-physical system models—like those for autonomous vehicles or power grids—that are composable, can be trained with machine learning, and come with mathematical guarantees of correctness, all within a single, scalable pipeline.