Feature augmentation based on information fusion rectification for few-shot image classification

Abstract

In the issue of few-shot image classification, due to lack of sufficient data, directly training the model will lead to overfitting. In order to alleviate this problem, more and more methods focus on non-parametric data augmentation, which uses the information of known data to construct non-parametric normal distribution to expand samples in the support set. However, there are some differences between base class data and new ones, and the distribution of different samples belonging to same class is also different. The sample features generated by the current methods may have some deviations. A new few-shot image classification algorithm is proposed on the basis of information fusion rectification (IFR), which adequately uses the relationship between the data (including the relationship between base class data and new ones, and the relationship between support set and query set in the new class data), to rectify the distribution of support set in the new class data. In the proposed algorithm, feature of support set is expanded through sampling from the rectified normal distribution, so as to augment the data. Compared with other image augmentation algorithms, the experimental results on three few-shot datasets show that the accuracy of the proposed IFR algorithm is improved by 0.93%-3.22%.

Event assigning based on hierarchical features and enhanced association for Chinese mayor's hotline

“…However, these ways often only focus on the dependency information of neighboring events on one side during one round of encoding, which is not effective to cope with the multiple events or multiple subjects overlapping. The literature 9 proposes trigger‐uniform sampling and confusion sampling ways to process datasets and introduces both adversarial training and triggered reconstruction techniques, which improve the model performance and enhance its generalization ability. Different from the above methods, this paper uses the attention mechanism and adds targeted design for Chinese special structure in the text feature extraction part to improve the accuracy of event classification in Chinese hotline events.…”

Defining a New NLP Playground

“…It is the worst of times. We are living in an incredibly exciting yet strange era of Natural Language Processing (NLP) research due to the recent advancements of large language models (LLMs) on various data modalities, from natural language (Brown et al, 2020) and programming language to vision (Radford et al, 2021;Li et al, 2022a;Wang et al, 2022b) and molecules (Edwards et al, 2022;Zeng et al, 2022;Su et al, 2022).…”