ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2019, Vol. 56 ›› Issue (12): 2521-2535.doi: 10.7544/issn1000-1239.2019.20190614

所属专题: 2019大数据知识工程及应用专题

• 其他应用技术 • 上一篇    下一篇

基于跨域对抗学习的零样本分类

刘欢1,郑庆华1,罗敏楠1,赵洪科2,肖阳3,吕彦章1   

  1. 1(西安交通大学计算机科学与技术学院 西安 710049);2(天津大学管理与经济学部 天津 300072);3(综合业务网国家重点实验室(西安电子科技大学) 西安 710071) (huanliucs@gmail.com)
  • 出版日期: 2019-12-01
  • 基金资助: 
    国家重点研发计划项目2018YFB1004500);国家自然科学基金面上项目(61572399);国家自然科学基金创新群体(61721002);教育部创新团队(IRT_17R86)

Cross-Domain Adversarial Learning for Zero-Shot Classification

Liu Huan1, Zheng Qinghua1, Luo Minnan1, Zhao Hongke2, Xiao Yang3, Lü Yanzhang1   

  1. 1(School of Computer Science and Technology, Xi’an Jiaotong University, Xi’an 710049);2(College of Management and Economics, Tianjin University, Tianjin 300072);3(State Key Laboratory of Integrated Services Networks (Xidian University), Xi’an 710071)
  • Online: 2019-12-01

摘要: 零样本学习旨在识别具有少量、甚至没有训练样本的未见类,这些类与可见类遵循不同的数据分布.最近,随着深度神经网络在跨模态生成方面的成功,使用合成的样本对未见数据进行分类取得了巨大突破.现有方法通过共享生成器和解码器,联合传统生成对抗网络和变分自编码器来实现样本的合成.然而,由于这2种生成网络产生的数据分布不同,联合模型合成的数据遵循复杂的多域分布.针对这个问题,提出跨域对抗生成网络(CrossD-AGN),将传统生成对抗网络和变分自编码器有机结合起来,基于类级语义信息为未见类合成样本,从而实现零样本分类.提出跨域对抗学习机制,引入2个对称的跨域判别器,通过判断合成样本属于生成器域分布还是解码器域分布,促使联合模型中的生成器/解码器不断优化,提高样本合成能力.在多个真实数据集上进行了广泛的实验,结果表明了所提出方法在零样本学习上的有效性和优越性.

关键词: 零样本学习, 生成模型, 跨模态生成, 跨域对抗学习, 联合模型

Abstract: Zero-shot learning (ZSL) aims to recognize novel categories, which have few or even no sample for training and follow a different distribution from seen classes. With the recent advances of deep neural networks on cross-modal generation, encouraging breakthroughs have been achieved on classifying unseen categories with their synthetic samples. Extant methods synthesize unseen samples with the combination of generative adversarial nets (GANs) and variational auto-encoder (VAE) by sharing the generator and the decoder. However, due to the different data distributions produced by these two kinds of generative models, fake samples synthesized by the joint model follow the complex multi-domain distribution instead of satisfying a single model distribution. To address this issue, in this paper we propose a cross-domain adversarial generative network (CrossD-AGN) to integrate the traditional GANs and VAE into a unified framework, which is able to generate unseen samples based on the class-level semantics for zero-shot classification. We propose two symmetric cross-domain discriminators along with the cross-domain adversarial learning mechanism to learn to determine whether a synthetic sample is from the generator-domain or the decoder-domain distribution, so as to drive the generator/decoder of the joint model to improve its capacity of synthesizing fake samples. Extensive experimental results over several real-world datasets demonstrate the effectiveness and superiority of the proposed model on zero-shot visual classification.

Key words: zero-shot learning, generative model, cross-modal generation, cross-domain adversarial learning, joint model

中图分类号: