Fast Rate Bounds for Multi-Task and Meta-Learning with Different Sample Sizes got accepted to NeurIPS 2025.