Training a model on multiple related tasks simultaneously so it learns shared patterns that improve performance across all tasks.