PARN: Position-Aware Relation Networks for Few-Shot Learning

Making machines learn like humans is the ultimate goal of artificial intelligence. Few-shot learning attempts to simulate the learning mechanism of humans, which is a task that can learn novel concepts from very few labeled samples. Due to the lack of sufficient labeled training data, existing methods often increase the risk of over-fitting or cause a considerable gap between clean and augmented data. To solve these problems, we present a novel compare network to perform robust few-shot learning in a metalearned end-to-end manner. Specifically, we argue that it is desirable to learn a robust encoder that can draw inferences about other cases from one example. To this end, we improve the accuracy of few-shot learning by mining the internal mechanism of deep networks, which can leverage label information more effectively. By introducing shift-invariant blocks and a self-attention block in our architecture, all these components are seamlessly integrated into our framework, which can give feedback to each other without data augmentation. Furthermore, we provide ablative analyses of different blocks to help understand how each term contributes to performance. Extensive experiments demonstrate that our method can provide more robust results and outperform state-of-the-art few-shot learning methods. INDEX TERMS Few-shot learning, shift-invariant, attention, compare network.

Section: Data Augmentation Based Methodsmentioning

confidence: 99%

“…The most relevant works of existing methods are [30] and [39]. However, they ignored the robustness of the network.…”

Section: Data Augmentation Based Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Robust Compare Network for Few-Shot Learning

Yang

Zhang

et al. 2020

“…PARN [22] PARN extended a deformable feature extractor, and designed dual correlation attention mechanisms to transform the original feature maps and generate the novel feature maps of position-aware.…”

Section: L-pn [25]mentioning

confidence: 99%

Feature Transformation Network for Few-Shot Learning

Wang

Zhou

2021

Few-shot learning researches to learn a novel concept from a handful of labeled samples. Due to the small amount of training data, deep network has the risk of over-fitting. Although many previous approaches based on metric criterion can make significant progress to tackle this challenge, they not only ignore the association between query set and support set when learning sample representation, but also fail to focus greater attention in the target area. To cope with these issues, we propose a novel feature transformation network (FTN) for few-shot image classification. Specifically, to draw inferences about other instances from only a few examples, it is expected to learn a model that has more discriminative representation of the target attributes and robust generalization ability. To this end, we introduce an attention-based affinity matrix to transform the semantical enhanced embedding vectors of query samples by associating the support set, thereby guiding the network to learn a sample representation that embodies higher semantic information in the target area. Furthermore, aiming at highlighting the object region in the feature maps, and strengthening the pertinence of similarity measurement between samples, a global and local feature fusion module is designed to fuse the support set samples features. The comprehensive experiments validate the doable of our model, and our method achieves the state-of-the-art performance on two public benchmark datasets, namely, general object dataset mini-ImageNet and fine-grained dataset Caltech-UCSD Birds-200-2011 (CUB). INDEX TERMS Few-shot learning, feature transformation, feature fusion, metric criterion.

“…It shows that when learning non-linear metric methods, with other network added, the training parameters in the model are increasing, it is easy for the network to overfit so as to cause a bad result. Besides, we compare our result with 17 state-of-the-art meta-metric learning algorithms, which are D-SVS [33], SN [34], SRPN [35], PML [29], DN4 [29], PCP [21], LCC [36], L2AE-D [37], IMP [23], PN [38], METRIC1 [39], DC [17], CovaMNET [40], AM3 [22], VFL [41], SHS [42], PARN [43]. As is shown in FIGURE 4, our results by combining Matching Network with fine-tuned ResNet50 outperform all of the state-of-the-art results, which illustrates the importance of transfer learning.…”

Section: )mentioning

confidence: 99%

Comparative Analysis on Classical Meta-Metric Models for Few-Shot Learning

Yang

Liu

2020

Few-shot learning are methods and scenarios learned from a small amount of labeled data. While recent meta-metric learning methods have made significant progress, there are still questions about what is the key point of these methods and how they work. To address these problems, in this paper, we evaluate the effects of different parts in classical models. To be specific, we 1) use four typical networks AlexNet, VGG16, GoogLeNet, and ResNet50 to replace the original feature extraction part in the Matching Network, Prototypical Network, and Relation Network, and compare the best results with 17 state-of-the-art meta-metric learning algorithms. 2) fix the feature extraction part of the Matching Network, Prototypical Network and Relation Network, and change the similarity measurement part of each into L1, L2, Cosine. 3) conduct above three models on datasets of different granularity. The experimental results show that for all models evaluated, the addition of non-pretrained networks will make the classification results worse, which shows that it is easy to overfit when using deep networks for few-shot learning. Changes in similarity measurement methods have a significant impact on results, which shows the importance to choose a suitable measurement. Moreover, there are differences in performance on different granularity datasets.