The Central Bank of Luxembourg has published Cahier d’études No. 196, a technical paper proposing a new machine-learning method to solve large-scale economic models, including frameworks with regime changes, using neural networks without backpropagation. The study situates the approach in the context of modelling challenges faced by central banks, where incorporating heterogeneity across households, firms and banks can sharply increase model dimensionality and complicate analysis of monetary policy transmission. It also links the dimensionality problem to demographic-transition analysis using overlapping generations (OLG) models, noting the Central Bank of Luxembourg’s LOLA model and the way model size grows with the number of generations when age cohorts are defined annually or quarterly. The paper describes regime changes as situations in which some households suddenly face credit constraints that limit consumption or investment over the cycle, and notes that the views expressed are those of the author and not necessarily those of the Central Bank of Luxembourg or the Eurosystem.