L2AE-D: Learning to Aggregate Embeddings for Few-shot Learning with Meta-level Dropout

NEUROCOMPUTING

卷:442

Article

Song, Heda¹  Torres, Mercedes Torres²  Ozcan, Ender¹  Triguero, Isaac¹ 

[1] Univ Nottingham, Sch Comp Sci, Computat Optimisat & Learning Lab, Nottingham NG8 1BB, England

[2] Univ Nottingham, Sch Comp Sci, Comp Vis Lab, Nottingham, England

关键词: Few-shot learning;    Meta-learning;    Metric-learning;    Embedding aggregation;    Attention mechanism;    Meta-level dropout;   

DOI  :  10.1016/j.neucom.2021.02.024

来源: Elsevier

【 摘 要 】

Few-shot learning focuses on learning a new visual concept with very limited labelled examples. A suc-cessful approach to tackle this problem is to compare the similarity between examples in a learned met-ric space based on convolutional neural networks. However, existing methods typically suffer from meta-level overfitting due to the limited amount of training tasks and do not normally consider the importance of the convolutional features of different examples within the same channel. To address these limitations, we make the following two contributions: (a) We propose a novel meta-learning approach for aggregat -ing useful convolutional features and suppressing noisy ones based on a channel-wise attention mecha-nism to improve class representations. The proposed model does not require fine-tuning and can be trained in an end-to-end manner. The main novelty lies in incorporating a shared weight generation module that learns to assign different weights to the feature maps of different examples within the same channel. (b) We also introduce a simple meta-level dropout technique that reduces meta-level overfitting in several few-shot learning approaches. In our experiments, we find that this simple technique signifi-cantly improves the performance of the proposed method as well as various state-of-the-art meta-learning algorithms. Applying our method to few-shot image recognition using Omniglot and miniImageNet datasets shows that it is capable of delivering a state-of-the-art classification performance. ? 2021 Elsevier B.V. All rights reserved.

【 授权许可】

Free