ALPINE is an ultra-lightweight spatial-relational architecture using only 22k-35k parameters for few-shot learning. It outperforms Prototypical Networks and MAML on CIFAR-FS and MiniImageNet while using up to 53% fewer parameters.
HOW THIS AFFECTS YOU
●
builderYou can deploy highly efficient few-shot classifiers on edge devices with minimal parameter budgets.
●
researcherThis demonstrates that content-adaptive patch localization can substitute for massive model scale in few-shot tasks.