ML-driven process optimization and mechanistic insights into controlled and living polymerizations

Controlled/living polymerization, owing to its excellent capability for regulating relative molecular mass and topological architecture, has become an important strategy for the synthesis of high-performance polymeric materials. However, such polymerization processes are typically characterized by multi-factor coupling, multiscale reaction dynamics, and highly nonlinear mappings between reaction conditions and polymer properties, which render conventional polymer design approaches reliant on empirical screening or mechanistic modeling limited by high computational cost, poor cross-system transferability, and restricted global optimization capability. In recent years, machine learning (ML) methods, leveraging multi-source data from experiments, literature reports, and computational simulations, have demonstrated strong capability in efficiently uncovering nonlinear relationships between reaction conditions, monomer/catalyst structures, and polymer properties. Even under limited data availability, ML enables rapid property prediction and inverse reaction design, providing a new paradigm for the rational design of controlled/living polymerization processes. This review summarizes recent advances in the application of ML to atom transfer radical polymerization, reversible addition-fragmentation chain transfer polymerization, ring-opening polymerization, and controlled/living polymerization, covering key aspects such as dataset construction, feature engineering, model training and optimization, as well as physical constraints and model interpretability. Furthermore, the integration of ML with high-throughput synthesis, online/in situ characterization techniques, and self-driving experimental platforms is discussed. Finally, in view of challenges including data scarcity, limited model interpretability, and insufficient automation, future perspectives are proposed, emphasizing the development of standardized polymerization databases, physically informed ML models, and self-driving experimental systems.