结合知识图谱的变分自编码器零样本图像识别

Variational Auto-Encoder Combined with Knowledge Graph Zero-Shot Learning

近年来,结合生成模型的零样本算法得到了广泛的研究,但此类方法通常仅使用属性注释,缺少类别语义,而单一信息对类别表征能力不够强,容易产生域偏移,影响知识迁移的效果,进而降低分类结果的准确率。为了解决此问题,提出一种结合知识图谱变分自编码器零样本识别算法(KG-VAE),通过构建联合类别分级结构,类别文本描述和词向量的层次结构化知识图谱作为语义信息库,将知识图谱中丰富的语义知识结合到以变分自编码器为基础的生成模型中,使生成的潜在特征更好保留有效的判定性信息,减小域偏移,促进知识迁移。在四个公开的零样本数据集上进行了实验,对比基准方法CADA-VAE,分类平均准确率有一定的提高;同时利用消融实验证明了知识图谱作为语义辅助信息的有效性。

Recently,zero-shot learning combine with generative model has been widely studied,but such methods usually only use attributes,lacking class semantics,a single information cannot strong enough to represent the class.It is easy to cause the domain-shift problem and affect knowledge transfer,the accuracy of classification results will be decreased.In order to solve this problem,variational auto-encoder combined with knowledge graph zero-shot learning(KG-VAE)is proposed.By building hierarchical structured knowledge graph with class description and word vector as the semantic information,the rich semantic information in the knowledge graph is combined into the VAE model.The generated potential features retain better effective deterministic information,it is effective to reduce domain shift and promote knowledge transfer.According to the evaluation with four public datasets,and compared with the baseline method CADA-VAE,the average classification accuracy is improved.At the same time,the availability of knowledge graph is proved by ablation experiment.