End-to-end learning trains a model to map raw or minimally processed inputs directly to desired outputs with fewer manually engineered intermediate steps.
It can simplify pipelines and improve performance when enough suitable data is available, but it may reduce interpretability and make troubleshooting harder.