Towards Enabling Meta-Learning from Target Models

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 $\mathit{support}$ sets and $\mathit{query}$ sets to train and evaluate the solvers respectively due to simplicity ($\mathcal{S}/\mathcal{Q}$ protocol). However, another evaluation method that assesses the discrepancy between the solver and a target model is short of research ($\mathcal{S}/\mathcal{T}$ protocol). $\mathcal{S}/\mathcal{T}$ protocol has unique advantages such as offering more informative supervision, but it is computationally expensive. This paper looks into this special evaluation method and takes a step towards putting it into practice. We find that with a small ratio of tasks armed with target models, classic meta-learning algorithms can be improved a lot without consuming many resources. Furthermore, we empirically verify the effectiveness of $\mathcal{S}/\mathcal{T}$ protocol in a typical application of meta-learning, $\mathit{i.e.}$, few-shot learning. In detail, after constructing target models by fine-tuning the pre-trained network on those hard tasks, we match the task-specific solvers to target models via knowledge distillation. Experiments demonstrate the superiority of our proposal.