Current Issue Cover

余悦, 陈楠, 成科扬(江苏大学)

摘 要
目的 跨域少样本学习的主要挑战在于,很难将源域的知识推广到未知的目标域中。最近的一些少样本学习模型试图通过在元训练过程中诱导图像多样化来解决这一问题。然而,其中一些模拟未知领域任务的方法效果有限,因为它们不能有效地模拟领域偏移,其生成的内容变化范围狭窄,难以从域偏移中学习到有效的域不变知识。为了提升少样本模型的跨域泛化能力,我们提出了基于不确定性增强的域感知网络框架(UEDA)。方法 基于不确定性增强的域感知框架从特征通道视角探索和提取其中可用于缓解领域偏移的关键知识。首先提出了一个不确定性特征增强方法,将特征的充分统计定值定义为服从高斯分布的概率表示,以源域充分统计量为分布中心建模不确定性分布。随后,从不确定性分布中生成有别于加性扰动的挑战性特征,从而挖掘不同域之间的共性知识;其次,提出了基于不确定性增强的域感知方法,将源特征和生成特征视为来自不同领域的信息,利用域鉴别器计算特征通道与领域信息的相关性,从而帮助模型挖掘领域之间的潜在关联并鉴别出其中的域因果信息用于学习。结果 实验使用了Mini-ImageNet、CUB、Plantae、EuroSAT和Cropdiseases共五个数据集来评估所提出方法的跨域泛化表现。实验遵从纯源域泛化,其中在GNN分类框架下,以Mini-ImageNet数据集作为源域,模型在后四个目标域的1-shot和5-shot设置下其平均精度分别为59.50%、47.48%、79.04%和75.08%,证明了所提出的方法能有效提高基于源域的跨域图像分类能力。结论 本文所提出的基于不确定性增强的域感知网络框架使得模型在训练阶段适应各种域偏移,并从中学习到有效的可泛化知识,从而提高在少样本条件下的跨域图像分类能力。
Uncertainty domain awareness network for cross-domain few-shot image classification

yuyue, chennan, chengkeyang(jiangsu university)

Objective Inspired by the fast learning properties of humans and transfer learning, researchers have proposed a new machine learning paradigm--few-shot learning. The purpose of few-shot learning is to enable models to quickly learn new knowledge or concepts from a limited amount of labelled data that can be used in unknown categories. Currently, few-shot image classification is based on the framework of meta-learning, which divides the model learning process into a meta-training phase and a meta-testing phase. Existing solutions can be broadly classified according to the differences in ideas:1) Optimisation-based methods, the basic idea of which is to allow the model to find the parameters that make the performance optimal under the training of multiple sets of task data; 2) Metric learning-based methods, whose core idea is to construct an optimal embedding space for measuring distances so that the distance between similar samples is as small as possible; 3) Data manipulation-based methods, which use some basic data augmentation (e.g., rotating, clipping, adding noise, etc.) to increase the diversity of the training samples and the amount of data in these three main categories. However, these works tend to follow strict assumptions such as smoothness assumption, clustering assumption and prevalence assumption, etc., and require that the training data and the test data come from under the same distribution. This makes it difficult to ensure data from the same distribution setting during the training process of the model on certain real-world scenarios, such as medical imaging, military applications, and finance, where issues such as difficulty in data access and data privacy make it challenging to use labelled data from other domains to provide a priori knowledge. Therefore, in order to alleviate the problem of domain distribution bias encountered in the learning process of the few-shot model, we propose a uncertainty enhancement-based domain-awareness network. Method Uncertainty enhancement-based domain-awareness approach explores and extracts key knowledge from the feature channel perspective in which can be used to mitigate domain bias. An uncertainty feature augmentation approach is first proposed, where the feature channel contains both domain-relevant and domain-independent information, suggesting that the generalisation learning of the model may be correlated with the feature channel"s ability to extract domain-generalised knowledge. However, most of the work treats feature statistics as deterministic and uses additive perturbations (i.e., swapping and interpolation) to achieve augmentation, practices that may lead to models that are negatively affected by too much domain-specific information. The uncertainty enhancement approach models the uncertainty distribution by defining the feature sufficient statistics fixed value as a probabilistic representation of uncertainty obeying a Gaussian distribution with the source feature sufficient statistic as the centre of the distribution and the standard deviation defined as being the potential range of variation of the probability. The new features generated by the uncertainty enhancement method can be effectively distinguished from the source domain features. The second part of UEDA is a domain-awareness approach. In the domain-aware module, we consider source and generated features as information from different domains, and ensure that both features are within a reasonable challenging offset by maximising the inter-domain differences. A domain discriminator is also introduced to compute the correlation between each channel and the domain information as a way to extract effective generalisable knowledge. Result We evaluate the cross-domain generalisation performance of the proposed method on a total of five datasets, Mini-ImageNet, CUB, Plantae, Cropdiseases and EuroSAT. The experiments follow single-domain generalisation, using the Mini-ImageNet dataset as the source domain and the latter four datasets as the target domains, and are compared with current mainstream work under three classification frameworks, namely MatchingNet, RelationNet and GNN. The experiments follow 5-way 1-shot and 5-way 5-shot settings on CUB, Plantae, EuroSAT, and Cropdiseases datasets, and the accuracies of the proposed UEDA are 41.01% and 58.78%, 38.07% and 51.36%, respectively, under the MatchingNet classifier, 58.37% and 80.45%, 59.48% and 69.91%, and under GNN classifier, the accuracy of UEDA is 49.36% and 69.65%, 38.48% and 56.49%, 68.98% and 89.11%, and 64.87% and 85.29%, respectively. Comparative experimental results demonstrate that the proposed UEDA method can effectively improve the cross-domain generalisation of the model. Furthermore, ablation experiments were conducted to validate the effectiveness of the modules of the proposed methodology, and the results show that the modules are mutually reinforcing and are indispensable in the overall methodology. Conclusion The uncertainty enhancement-based domain-awareness network proposed in this paper allows the model to adapt to various domain offsets during the training phase and learn effective generalisable knowledge from them, thus improving cross-domain image classification under fewer samples.