LWR-Net: Learning without retraining for scalable multi-task adaptation and domain-agnostic generalisation | Publicación