Behavior regularized prototypical networks for semi-supervised few-shot image classification

作者：

Highlights：

• We propose a Behavior Regularized Prototypical Network (BR-ProtoNet) for few-shot image classification in semi-supervised scenarios.

• BR-ProtoNet enables metric learning to benefit from readily-available unlabeled data.

• We construct complementary constraints to regularize the model’s behavior over the neighborhoods of training instances and along the interpolation paths among them.

• The constructed regularization encourages the learnt embedding space to possess the property of proximity preservation.

摘要

•We propose a Behavior Regularized Prototypical Network (BR-ProtoNet) for few-shot image classification in semi-supervised scenarios.•BR-ProtoNet enables metric learning to benefit from readily-available unlabeled data.•We construct complementary constraints to regularize the model’s behavior over the neighborhoods of training instances and along the interpolation paths among them.•The constructed regularization encourages the learnt embedding space to possess the property of proximity preservation.

论文关键词：Few-shot learning,Semi-supervised learning,Image classification,Prototypical networks

论文评审过程：Received 23 March 2020, Revised 21 September 2020, Accepted 21 November 2020, Available online 5 December 2020, Version of Record 14 December 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107765