Few-sample learning method based on sample-level attention network

Abstract

The invention relates to a few-sample learning method based on a sample-level attention network. The method comprises the steps of respectively inputting a sample of a support set and a sample of a query set into a feature extraction module, and obtaining a corresponding support set feature vector and a query set feature vector; inputting the support set feature vector corresponding to each classinto a sample-level attention network module to obtain a class prototype of each class; calculating the distance between the query set feature vector and each class prototype to obtain the probabilitydistribution of the class to which the query set feature vector belongs; jointly training the feature extraction module and the sample-level attention network module by adopting cross entropy loss ofthe query set and classification loss of the support set, and obtaining a gradient updating network through back propagation. According to the method, a new target task can be solved only by learninga large number of similar learning tasks, and for the target task, model updating is not needed any more.

Description

technical field [0001] The invention relates to the technical field of sample image classification, in particular to a few-sample learning method based on a sample-level attention network. Background technique [0002] In recent years, with the help of deep networks, computers have surpassed humans in image recognition tasks, but deep networks also have three problems: First, training deep network models requires a large amount of labeled data, so most image recognition technologies are Focus on object recognition with large-scale datasets. Second, in many scenarios in reality, there is no large amount of labeled data, or a lot of labeled data needs to be labeled by domain experts, and the cost of obtaining large-scale labeled datasets is too high, which makes it impossible to use deep learning in many fields. Third, deep learning requires a large amount of data and multiple iterations to train a model, but when faced with the emergence of new categories, it needs to be ret...

AI Technical Summary

Problems solved by technology

[0002] In recent years, with the help of deep networks, computers have surpassed humans in image recognition tasks, but deep networks also have three problems: First, training deep network models requires a large amount of labeled data, so most image recognition technologies are Focus on object recognition with large-scale datasets

Second, in many scenarios in reality, there is no large amount of labeled data, or a lot of labeled data requires domain experts to label, and the cost of obtaining large-scale labeled data sets is too high, which makes it impossible to use deep learning in many fields.

Third, deep learning requires a large amount of data and multiple iterations to train a model, but when faced with the emergence of new categories, it needs to be retrained

[0008] However, the above two methods have disadvantages, that is, the network needs to be fine-tuned when facing new few-sample learning tasks.

Images