Meta-prediction Model for Distillation-Aware NAS on Unseen Datasets

Distillation-aware Network Architecture Search (DaNAS) aims to search for an optimal student architecture that can obtain the best performance and/or efficiency when distilling the knowledge from a given teacher model. Previous DaNAS methods have mostly tackled the search for the network architecture for a fixed source/target tasks and the teacher, which are not generalized well on a new task, thus need to perform costly search for any new combination of the domains and the teachers. For standard NAS tasks without KD, meta-learning-based computationally efficient NAS methods have been proposed, which learn the generalized search process over multiple tasks and transfer the knowledge obtained over those tasks to a new task. However, since they assume learning from scratch without KD from a teacher, they might not be ideal for DaNAS scenarios, which could significantly affect the final accuracies of the architectures obtained from the search. To eliminate excessive computational cost of DaNAS methods and the sub-optimality of rapid NAS methods, we propose a distillation-aware meta accuracy prediction model which can predict a given architecture's final performances on a dataset when performing KD with a given teacher, without having to actually train it on the target task. The experimental results demonstrate that our proposed meta-prediction model successfully generalizes to multiple unseen datasets for DaNAS tasks, largely outperforming existing meta-NAS methods and rapid NAS baselines.