Meta-learning can extract an inductive bias from previous learning experience and assist the training processes of new tasks. It is often realized through optimizing a meta-model with the evaluation loss of a series of task-specific solvers. Most existing algorithms sample non-overlapping 𝑠𝑢𝑝𝑝𝑜𝑟𝑡 sets and 𝑞𝑢𝑒𝑟𝑦 sets to train and evaluate the solvers respectively due to simplicity (𝒮/𝒬 protocol). However, another evaluation method that assesses the discrepancy between the solver and…