Current few-shot learning research often assumes predefined task configurations and evaluates under fixed N-way K-shot settings. In realistic deployments, the target configuration is unknown at ...